Write
research papers
10x faster, all in one free app.

A free, local-first workspace for serious research writing. Draft, compile, cite, and review in LaTeX, Typst, or Markdown. Use AI that can edit your project, compile it, and check the result. Your files stay on your disk, with every version tracked in Git.

Download for free Read the docs

OleaflyEdgeSense: Energy Aware Wi…

Compile

prajwal-svm

Source Tree

.gitignore

main.tex

project.json

README.md

references.bib

refs.bib

Structure 38

Filter project map…

main.texmain.tex:19:1

Introductionmain.tex:19:1

System designmain.tex:26:1

fig:architecturemain.tex:43:8

System designmain.tex:26:1

Deployment and evaluationmain.tex:45:1

tab:resultsmain.tex:56:69

Deployment and evaluationmain.tex:45:1

equationmain.tex:50:1

Resultsmain.tex:65:1

figuremain.tex:71:1

Ablations, safety, and limitat…main.tex:81:1

Conclusionmain.tex:87:1

fairmain.tex:92:1

tinymlmain.tex:93:1

main.tex

1\documentclass[conference]{IEEEtran}

2\usepackage{amsmath,amssymb,booktabs,graphicx,xcolor,tikz}

3\usepackage[hidelinks]{hyperref}

4\usetikzlibrary{positioning,arrows.meta,shapes.geometric}

5\title{EdgeSense: Energy-Aware Wildfire Detection with Collaborative Sensor Networks}

6\author{\IEEEauthorblockN{Nadia Flores, Min-jun Lee, and Caleb Morgan}

7\IEEEauthorblockA{Center for Resilient Computing, Cascadia Institute of Technology\\

8Seattle, WA, USA \quad \{nflores,mjlee,cmorgan\}@example.edu}}

9\begin{document}\maketitle

10\begin{abstract}

11We present EdgeSense, a low-power wildfire early-warning system that fuses thermal,

12particulate, humidity, and wind observations across a mesh of battery-powered

13nodes. A gated temporal model runs locally and transmits only uncertain windows~\cite{lecun2015deep}.

14In a 16-week controlled field deployment, EdgeSense detected 94.1\% of burn events

15at 0.7 false alarms per node-week while reducing radio energy by 63\% relative to

16periodic streaming. Results use a synthetic but realistic demo dataset.

17\end{abstract}

18\begin{IEEEkeywords}edge computing, environmental sensing, wildfire, TinyML, sensor networks\end{IEEEkeywords}

19\section{Introduction}

20Wildfire ignition can develop faster than sparse satellite revisit cycles and

21manual reports~\cite{hastie2009}. Dense ground networks provide local measurements but create a

22difficult systems trade-off: radios consume most of a node's energy, yet delaying

23communication can hide the earliest evidence~\cite{wilkinson2016}. EdgeSense moves triage to the node

24and treats transmission as a decision under uncertainty.

26\section{System design}

27Each node samples temperature, relative humidity, particulate matter, and wind

28once per minute~\cite{vaswani2017}. A 12\,kB quantized temporal convolutional network scores

29five-minute windows~\cite{gelman2013}. Scores below 0.25 are stored locally; scores between 0.25 and

300.70 trigger peer confirmation; scores above 0.70 trigger an immediate gateway

31packet.

32\begin{figure}[t]\centering

33\begin{tikzpicture}[font=\scriptsize,node distance=4mm,

34n/.style={draw,rounded corners,fill=blue!6,minimum width=18mm,minimum height=7mm},

35a/.style={-{Stealth[length=1.7mm]},thick}]

36\node[n] (s) {Sensors}; \node[n,right=of s] (f) {Feature buffer};

37\node[n,right=of f] (m) {Tiny model}; \node[n,below=of m] (p) {Peer vote};

38\node[n,left=of p] (g) {Gateway};

39\draw[a](s)--(f);\draw[a](f)--(m);\draw[a](m)--(p);\draw[a](p)--(g);

40\draw[a,bend left=25](m) to node[above]{high confidence}(g);

41\end{tikzpicture}

42\caption{EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.}

43\label{fig:architecture}\end{figure}

45\section{Deployment and evaluation}

46We deployed 48 nodes across three ecological zones~\cite{wilkinson2016}. Controlled burns and heated

47aerosol challenges generated 186 positive windows; 11,420 background windows

48captured fog, dust, vehicle exhaust, and sensor maintenance~\cite{lundberg2017}. Splits were grouped

49by day to prevent temporal leakage.

50\begin{equation}

51\mathcal{L}=-w_+y\log p-w_-(1-y)\log(1-p)+\lambda\lVert\theta\rVert_2^2.

52\end{equation}

53We compare periodic streaming, fixed thresholds, an ungated neural baseline, and

54EdgeSense. The primary metrics are event recall, false alarms per node-week,

55median alert latency, and measured radio energy.

56\begin{table}[t]\centering\caption{Field evaluation results.}\label{tab:results}

57\begin{tabular}{@{}lrrrr@{}}\toprule

58Method & Recall & FA/wk & Latency & Energy\\\midrule

59Periodic stream & .957 & 1.8 & 72 s & 100\%\\

60Fixed threshold & .876 & 3.4 & 61 s & 41\%\\

61Ungated TCN & .946 & 1.1 & 58 s & 52\%\\

62\textbf{EdgeSense} & \textbf{.941} & \textbf{0.7} & \textbf{54 s} & \textbf{37\%}\\\bottomrule

63\end{tabular}\end{table}

65\section{Results}

66EdgeSense retains 98.3\% of the streaming baseline's recall while using 37\% of

67its radio energy~\cite{gelman2013}. Peer voting removes 31\% of fog-related false alarms~\cite{hastie2009}. At the

68median observed solar input, the modeled service interval rises from 11 to 29

69months. Performance degrades gracefully when one sensing channel is unavailable.

71\begin{figure}[t]\centering

72\begin{tikzpicture}[x=.72cm,y=.55cm,font=\scriptsize]

73\draw[->](0,0)--(7,0)node[right]{week};\draw[->](0,0)--(0,5)node[above]{recall};

74\draw[blue,very thick] plot coordinates{(0,.8)(1,1.6)(2,2.2)(3,2.8)(4,3.5)(5,4.0)(6,4.25)};

75\draw[orange,very thick,dashed] plot coordinates{(0,.7)(1,1.3)(2,1.9)(3,2.4)(4,2.8)(5,3.1)(6,3.25)};

76\node[blue] at (5.2,4.5){EdgeSense};\node[orange] at (5,2.6){threshold};

77\end{tikzpicture}

78\caption{Cumulative detection performance remained stable across the deployment.}

79\end{figure}

81\section{Ablations, safety, and limitations}

82Removing peer voting increases false alarms by 44\%; removing wind direction

83reduces recall by 3.8 points~\cite{lundberg2017}. EdgeSense is advisory: alerts require confirmation

84by the incident-management system, and nodes never autonomously dispatch a public

85warning~\cite{vaswani2017}. The deployment did not cover crown fires or winter conditions.

87\section{Conclusion}

88Selective communication can extend unattended sensing without hiding early

89signals~\cite{hastie2009}. The next deployment will test transfer across biomes and signed

90over-the-air model updates.

91\begin{thebibliography}{2}

92\bibitem{fair} M~\cite{wilkinson2016}. Wilkinson et al., \emph{The FAIR Guiding Principles}, \emph{Scientific Data}, 2016.

93\bibitem{tinyml} P. Warden and D. Situnayake, \emph{TinyML}. O'Reilly, 2019.

94\end{thebibliography}

95\end{document}

Logs 1.9s 100% 1 of 2

Compiling 1/3 done

1 running · 1 queued

Language analysisCurrent project revision analyzed
CompilingProducing and verifying PDF output
Rendering PDFWaiting for verified compile output

EdgeSense: Energy-Aware Wildfire Detection with
Collaborative Sensor Networks

Nadia Flores, Min-jun Lee, and Caleb Morgan
Center for Resilient Computing, Cascadia Institute of Technology
Seattle, WA, USA {nflores,mjlee,cmorgan}@example.edu

Fig. 1. EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.

Abstract—We present EdgeSense, a low-power wildfire early-warning system that fuses thermal, particulate, humidity, and wind observations across a mesh of battery-powered nodes. A gated temporal model runs locally and transmits only uncertain windows. In a 16-week controlled field deployment, EdgeSense detected 94.1% of burn events at 0.7 false alarms per node-week while reducing radio energy by 63% relative to periodic streaming. Results use a synthetic but realistic demo dataset.

Index Terms—edge computing, environmental sensing, wildfire, TinyML, sensor networks

I. INTRODUCTION

Wildfire ignition can develop faster than sparse satellite revisit cycles and manual reports . Dense ground networks provide local measurements but create a difficult systems trade-off: radios consume most of a node's energy, yet delaying communication can hide the earliest evidence . EdgeSense moves triage to the node and treats transmission as a decision under uncertainty.

II. SYSTEM DESIGN

Each node samples temperature, relative humidity, particulate matter, and wind once per minute . A 12 kB quantized temporal convolutional network scores five-minute windows . Scores below 0.25 are stored locally; scores between 0.25 and 0.70 trigger peer confirmation; scores above 0.70 trigger an immediate gateway packet .

III. DEPLOYMENT AND EVALUATION

We deployed 48 nodes across three ecological zones . Controlled burns and heated aerosol challenges generated 186 positive windows; 11,420 background windows captured fog, dust, vehicle exhaust, and sensor maintenance . Splits were grouped by day to prevent temporal leakage.

ℒ = −w₊y log p − w₋(1−y) log(1−p) + λ∥θ∥²₂.(1)

TABLE I
Field evaluation results.

Method	Recall	FA/wk	Latency	Energy
Periodic stream	.957	1.8	72 s	100%
Fixed threshold	.876	3.4	61 s	41%
Ungated TCN	.946	1.1	58 s	52%
EdgeSense	.941	0.7	54 s	37%

Fig. 2. Cumulative detection performance remained stable across the deployment.

We compare periodic streaming, fixed thresholds, an ungated neural baseline, and EdgeSense. The primary metrics are event recall, false alarms per node-week, median alert latency, and measured radio energy.

IV. RESULTS

EdgeSense retains 98.3% of the streaming baseline's recall while using 37% of its radio energy . Peer voting removes 31% of fog-related false alarms . At the median observed solar input, the modeled service interval rises from 11 to 29 months. Performance degrades gracefully when one sensing channel is unavailable.

V. ABLATIONS, SAFETY, AND LIMITATIONS

Removing peer voting increases false alarms by 44%; removing wind direction reduces recall by 3.8 points . EdgeSense is advisory: alerts require confirmation by the incident-management system, and nodes never autonomously dispatch a public warning . The deployment did not cover crown fires or winter conditions.

VI. CONCLUSION

Selective communication can extend unattended sensing without hiding early signals . The next deployment will test transfer across biomes and signed over-the-air model updates.

REFERENCES

[1]L. Giglio, J. Descloitres, C. O. Justice, and Y. J. Kaufman, “An enhanced contextual fire detection algorithm for MODIS,” Remote Sensing of Environment, vol. 87, no. 2–3, pp. 273–282, 2003.

[2]W. Schroeder, P. Oliva, L. Giglio, and I. A. Csiszar, “The new VIIRS 375 m active fire detection data product,” Remote Sensing of Environment, vol. 143, pp. 85–96, 2014.

[3]U. Dampage, L. Bandaranayake, R. Wanasinghe, K. Kottahachchi, and B. Jayasanka, “Forest fire detection system using wireless sensor networks and machine learning,” Scientific Reports, vol. 12, no. 1, p. 46, 2022.

[4]K. S. Adu-Manu, N. Adam, C. Tapparello, H. Ayatollahi, and W. Heinzelman, “Energy-harvesting wireless sensor networks: A review,” ACM Transactions on Sensor Networks, vol. 14, no. 2, pp. 1–50, 2018.

[5]F. Adelantado, X. Vilajosana, P. Tuset-Peiro, B. Martinez, J. Melia-Segui, and T. Watteyne, “Understanding the limits of LoRaWAN,” IEEE Communications Magazine, vol. 55, no. 9, pp. 34–40, 2017.

[6]S. Bai, J. Z. Kolter, and V. Koltun, “An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,” arXiv:1803.01271, 2018.

[7]B. Jacob, S. Kligys, B. Chen, M. Zhu, M. Tang, A. Howard, H. Adam, and D. Kalenichenko, “Quantization and training of neural networks for efficient integer-arithmetic-only inference,” in Proc. IEEE/CVF CVPR, 2018, pp. 2704–2713.

[8]P. P. Ray, “A review on TinyML: State-of-the-art and prospects,” Journal of King Saud University - Computer and Information Sciences, vol. 34, no. 4, pp. 1595–1623, 2022.

[9]Y. Geifman and R. El-Yaniv, “Selective classification for deep neural networks,” arXiv:1705.08500, 2017.

Zoom in

Zoom out

Fit to width

Fit to height

Reset to 100%

25%

50%

75%

100%

150%

200%

400%

Undefined reference

bibentry · wilkinson2016

@article
@article{wilkinson2016,
references.bib:11

Plan

Inventory citations used in main.tex and references.bib

Search literature for papers matching each cited claim

Verify every candidate citation is real

Add verified BibTeX entries to references.bib

Replace unresolved cites in main.tex

Recompile and confirm all citations resolve

Connect an AI provider to continue

Bring your own API key (OpenAI, Anthropic, Groq, and more) or run a model locally with Ollama. The assistant can read and edit files, compile your project, and verify the PDF.

Connect a providerRun a local model with Ollama

How can I help with your research?

Working on "EdgeSense: Energy Aware Wildfire Detection"

Find papers to cite Write a literature review Fix any source errors in my document Create a new section called 'Publications' Find every citation in the project Recompile and check for errors

Import your Zotero or EndNote library

Find papers to cite

Thinking...

Thought for 10s

update_todos

read_file

I'll inspect the current citations first, then search for papers that actually support the claims.

Thinking...

I need sources that match the paper's actual claims:
- satellite fire detection
- low-power sensor networks
- TinyML and quantized inference

I'll keep only results with verifiable DOI metadata.

literature_search

alphaxiv_search

Quick audit: three citation keys do not resolve, and two existing entries point to unrelated papers. I'm searching for better matches on wildfire sensing, TinyML, and low-power networks.

literature_search

verify_citation

Four solid matches so far: MODIS fire detection, a field-tested sensor-network detector, an energy-harvesting survey, and a TinyML review. I'm checking the DOI metadata before I touch the bibliography.

verify_citation

update_todos

The citations check out. I'll add nine verified entries to references.bib, then place them beside the claims they support.

write_fileApproved

The bibliography is in place. Next I'm replacing the unresolved keys in main.tex with the verified sources.

replace_in_fileAlways allowed

compile

verify_pdf_pages

Done. I added nine verified references, replaced the three unresolved keys, and recompiled the paper. The two-page PDF now has a complete IEEE bibliography with no missing citations.

Restore code to before this response

The assistant wants to change your files

write_filereferences.bibline 1

references.bibfirst change · L1

1@article{open_science2015,

2 title = {Estimating the Reproducibility of Psychological Science},

1@article{giglio2003,

2 author = {Giglio, Louis and Descloitres, Jacques and Justice, Christopher O.},

3 title = {An Enhanced Contextual Fire Detection Algorithm for {MODIS}},

4 journal = {Remote Sensing of Environment},

5 year = {2003}, doi = {10.1016/S0034-4257(03)00184-6}

6}

Reject Always allow Approve

The assistant wants to change your files

replace_in_filemain.texline 21

main.texfirst change · L21

20Wildfire ignition can develop faster than sparse satellite revisit cycles and

21manual reports~\cite{hastie2009}. Dense ground networks provide local measurements

21manual reports~\cite{giglio2003,schroeder2014}. Dense ground networks provide local measurements

22but create a difficult systems trade-off: radios consume most of a node's energy,

23yet delayed communication can hide the earliest evidence~\cite{wilkinson2016}.

23yet delayed communication can hide the earliest evidence~\cite{dampage2022,adumanu2018}.

Reject Always allow Approve

Ask AI to help with your document...

Prompts gpt-5.6-sol

Logs 4.3s 105% 1of 2

EdgeSense: Energy-Aware Wildfire Detection with
Collaborative Sensor Networks

Nadia Flores, Min-jun Lee, and Caleb Morgan
Center for Resilient Computing, Cascadia Institute of Technology
Seattle, WA, USA {nflores,mjlee,cmorgan}@example.edu

Fig. 1. EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.

Index Terms—edge computing, environmental sensing, wildfire, TinyML, sensor networks

I. INTRODUCTION

II. SYSTEM DESIGN

III. DEPLOYMENT AND EVALUATION

ℒ = −w₊y log p − w₋(1−y) log(1−p) + λ∥θ∥²₂.(1)

TABLE I
Field evaluation results.

Method	Recall	FA/wk	Latency	Energy
Periodic stream	.957	1.8	72 s	100%
Fixed threshold	.876	3.4	61 s	41%
Ungated TCN	.946	1.1	58 s	52%
EdgeSense	.941	0.7	54 s	37%

Fig. 2. Cumulative detection performance remained stable across the deployment.

IV. RESULTS

V. ABLATIONS, SAFETY, AND LIMITATIONS

VI. CONCLUSION

Selective communication can extend unattended sensing without hiding early signals . The next deployment will test transfer across biomes and signed over-the-air model updates.

REFERENCES

[2]W. Schroeder, P. Oliva, L. Giglio, and I. A. Csiszar, “The new VIIRS 375 m active fire detection data product,” Remote Sensing of Environment, vol. 143, pp. 85–96, 2014.

[6]S. Bai, J. Z. Kolter, and V. Koltun, “An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,” arXiv:1803.01271, 2018.

[8]P. P. Ray, “A review on TinyML: State-of-the-art and prospects,” Journal of King Saud University - Computer and Information Sciences, vol. 34, no. 4, pp. 1595–1623, 2022.

[9]Y. Geifman and R. El-Yaniv, “Selective classification for deep neural networks,” arXiv:1705.08500, 2017.

Stale

references.bib

1@article{giglio2003,

2 title = {An Enhanced Contextual Fire Detection Algorithm for MODIS},

3 year = {2003},

4}

6@article{schroeder2014,

7 title = {The New VIIRS 375 m Active Fire Detection Data Product},

8 year = {2014},

9}

11@article{dampage2022,

12 title = {Forest Fire Detection System Using Wireless Sensor Networks},

13 year = {2022},

14}

16@article{adumanu2018,

17 title = {Energy-Harvesting Wireless Sensor Networks: A Review},

18 year = {2018},

19}

21@article{adelantado2017,

22 title = {Understanding the Limits of LoRaWAN},

23 year = {2017},

24}

26@article{bai2018,

27 title = {An Empirical Evaluation of Generic Convolutional Networks},

28 year = {2018},

29}

31@article{jacob2018,

32 title = {Quantization and Training of Neural Networks},

33 year = {2018},

34}

36@article{ray2022,

37 title = {A Review on TinyML},

38 year = {2022},

39}

41@article{geifman2017,

42 title = {Selective Classification for Deep Neural Networks},

43 year = {2017},

44}

□▭○◯◇▱TΣ

Drag Shapes to Move · Drag Handles to Connect · Space + Drag to Pan · Double-Click to Edit Text

＋−⌗

Output Probabilities

Softmax

Linear

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

N×

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

Add & Norm

Masked Multi-Head
Attention

N×

Positional Encoding

Input Embedding

Inputs

Positional Encoding

Output Embedding

Outputs (shifted right)

Snippets　□　○　→　—　{ }

1\definecolor{c0F172A}{HTML}{0F172A}

2\definecolor{c1E293B}{HTML}{1E293B}

3\definecolor{cC7C7CC}{HTML}{C7C7CC}

4\definecolor{cCDEDD0}{HTML}{CDEDD0}

5\definecolor{cCFE8F8}{HTML}{CFE8F8}

6\definecolor{cDAD2F0}{HTML}{DAD2F0}

7\definecolor{cE9E9EC}{HTML}{E9E9EC}

8\definecolor{cEEF2C3}{HTML}{EEF2C3}

9\definecolor{cF9D7D9}{HTML}{F9D7D9}

10\definecolor{cFDE3C7}{HTML}{FDE3C7}

11\definecolor{cFFFFFF}{HTML}{FFFFFF}

12\begin{tikzpicture}[>={Triangle[length=0.313cm,width=0.313cm]}]

  \node (n1_mds37) at (6.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Inputs};

  \node (n2_hnw3j) at (6.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Input Embedding};

  \node (n3_x9vjf) at (4.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {};

  \node (n4_aa0b6) at (6.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+};

  \node (n5_lbeum) at (2.625,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding};

  \node (n6_gr85r) at (6.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention};

  \node (n7_it3d0) at (6.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n8_0ae9p) at (6.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward};

  \node (n9_31jhg) at (6.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n11_3mv5b) at (1.875,-13.575) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×};

  \node (n12_y6hil) at (16.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Outputs (shifted right)};

  \node (n13_oukbf) at (16.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Output Embedding};

  \node (n14_e6m10) at (16.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+};

  \node (n15_te7fn) at (18.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {};

  \node (n16_otz7n) at (19.875,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding};

  \node (n17_nw4kz) at (16.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Masked Multi-Head Attention};

  \node (n18_feep8) at (16.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n19_9lw0g) at (16.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention};

  \node (n20_lz6h6) at (16.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n21_iyi8d) at (16.25,-8.825) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward};

  \node (n22_hapki) at (16.25,-7.075) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n24_qwbl3) at (19.375,-11.825) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×};

  \node (n25_isnp1) at (16.25,-4.8) [draw=c1E293B, line width=0.025cm, fill=cDAD2F0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Linear};

  \node (n26_ggail) at (16.25,-2.8) [draw=c1E293B, line width=0.025cm, fill=cCDEDD0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Softmax};

  \node (n27_76g5v) at (16.25,-0.875) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.5cm, minimum height=1.25cm] {Output Probabilities};

38 \begin{scope}[on background layer]

    \node (n10_ma9el) at (6.25,-13.575) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=7.65cm] {};

    \node (n23_egiii) at (16.25,-11.7) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=11.4cm] {};

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-21.25) -- (6.25,-20.75) -- (6.25,-21.15) -- (6.25,-20.65);

42 \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-19.5) -- (6.25,-19) -- (6.25,-19.1) -- (6.25,-18.6);

43 \draw[->, line width=0.05cm] (n3_x9vjf.east) to[out=0, in=180] (n4_aa0b6.west);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-17.75) -- (6.25,-17.25) -- (6.25,-17.4) -- (6.25,-16.9);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-15.75) -- (6.25,-15.25) -- (6.25,-15.65) -- (6.25,-15.15);

46 \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-14) -- (6.25,-13.5) -- (6.25,-13.65) -- (6.25,-13.15);

47 \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-12) -- (6.25,-11.5) -- (6.25,-11.9) -- (6.25,-11.4);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-21.25) -- (16.25,-20.75) -- (16.25,-21.15) -- (16.25,-20.65);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-19.5) -- (16.25,-19) -- (16.25,-19.1) -- (16.25,-18.6);

50 \draw[->, line width=0.05cm] (n15_te7fn.west) to[out=180, in=0] (n14_e6m10.east);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-17.75) -- (16.25,-17.25) -- (16.25,-17.4) -- (16.25,-16.9);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-15.75) -- (16.25,-15.25) -- (16.25,-15.65) -- (16.25,-15.15);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-14) -- (16.25,-13.5) -- (16.25,-13.65) -- (16.25,-13.15);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-12) -- (16.25,-11.5) -- (16.25,-11.9) -- (16.25,-11.4);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-10.25) -- (16.25,-9.75) -- (16.25,-9.9) -- (16.25,-9.4);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-8.25) -- (16.25,-7.75) -- (16.25,-8.15) -- (16.25,-7.65);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-6.5) -- (16.25,-6) -- (16.25,-5.925) -- (16.25,-5.85) -- (16.25,-5.35);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-4.25) -- (16.25,-3.75) -- (16.25,-3.85) -- (16.25,-3.35);

59 \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-2.25) -- (16.25,-1.75) -- (16.25,-2) -- (16.25,-1.5);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (8.375,-10.825) -- (8.875,-10.825) -- (11.25,-10.825) -- (11.25,-12.575) -- (13.625,-12.575) -- (14.125,-12.575);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (5.825,-18.175) -- (5.325,-18.175) -- (3.625,-18.175) -- (3.625,-14.575) -- (4.125,-14.575);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (4.125,-14.575) -- (4.1,-14.575) -- (3.625,-14.575) -- (3.625,-10.825) -- (4.125,-10.825);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.675,-18.175) -- (17.175,-18.175) -- (18.875,-18.175) -- (18.875,-14.575) -- (18.375,-14.575);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-14.575) -- (18.875,-14.575) -- (18.875,-10.825) -- (18.4,-10.825) -- (18.375,-10.825);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-10.825) -- (18.875,-10.825) -- (18.875,-7.075) -- (18.4,-7.075) -- (18.375,-7.075);

66 \end{scope}

67\end{tikzpicture}

PREVIEWPNG scale2x　⌄Background

Output Probabilities

Softmax

Linear

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

N×

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

Add & Norm

Masked Multi-Head
Attention

N×

Positional Encoding

Input Embedding

Inputs

Positional Encoding

Output Embedding

Outputs (shifted right)

Compiling...

Import 6

↕ TECTONICdiagram

DiagramUpdated today

↕ TECTONICCatalyst Screening Data Report

DocumentUpdated Jul 15

↕ TECTONICAlex Chen - Google Resume

ResumeUpdated Jul 15

↕ TECTONICSleep and Memory Consolidation — Evidence Map

DocumentUpdated Jul 14

↕ TECTONICEdgeSense: Energy Aware Wildfire Detection

DocumentUpdated Jul 13

↕ TECTONICNeural Operators for Parametric PDEs

DocumentUpdated Jul 13

↕ TECTONICTheo Martins — HCI Researcher Resume

ResumeUpdated Jul 12

↕ TECTONICieee-journal

DocumentUpdated Jul 12

Oleafly›diagram

◉　prajwal-svm

main.tex　×

↶　↷　T⌄　 B　I　U　 ‹›　🔗　@　◇　✳　❞　▧　▦　⋯

ⓘ　⌕

Snippets　□　○　→　—　{ }

1\definecolor{c0F172A}{HTML}{0F172A}

2\definecolor{c1E293B}{HTML}{1E293B}

3\definecolor{cC7C7CC}{HTML}{C7C7CC}

4\definecolor{cCDEDD0}{HTML}{CDEDD0}

5\definecolor{cCFE8F8}{HTML}{CFE8F8}

6\definecolor{cDAD2F0}{HTML}{DAD2F0}

7\definecolor{cE9E9EC}{HTML}{E9E9EC}

8\definecolor{cEEF2C3}{HTML}{EEF2C3}

9\definecolor{cF9D7D9}{HTML}{F9D7D9}

10\definecolor{cFDE3C7}{HTML}{FDE3C7}

11\definecolor{cFFFFFF}{HTML}{FFFFFF}

12\begin{tikzpicture}[>={Triangle[length=0.313cm,width=0.313cm]}]

  \node (n1_mds37) at (6.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Inputs};

  \node (n2_hnw3j) at (6.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Input Embedding};

  \node (n3_x9vjf) at (4.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {};

  \node (n4_aa0b6) at (6.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+};

  \node (n5_lbeum) at (2.625,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding};

  \node (n6_gr85r) at (6.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention};

  \node (n7_it3d0) at (6.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n8_0ae9p) at (6.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward};

  \node (n9_31jhg) at (6.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n11_3mv5b) at (1.875,-13.575) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×};

  \node (n12_y6hil) at (16.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Outputs (shifted right)};

  \node (n13_oukbf) at (16.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Output Embedding};

  \node (n14_e6m10) at (16.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+};

  \node (n15_te7fn) at (18.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {};

  \node (n16_otz7n) at (19.875,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding};

  \node (n17_nw4kz) at (16.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Masked Multi-Head Attention};

  \node (n18_feep8) at (16.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n19_9lw0g) at (16.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention};

  \node (n20_lz6h6) at (16.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n21_iyi8d) at (16.25,-8.825) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward};

  \node (n22_hapki) at (16.25,-7.075) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm};

  \node (n24_qwbl3) at (19.375,-11.825) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×};

  \node (n25_isnp1) at (16.25,-4.8) [draw=c1E293B, line width=0.025cm, fill=cDAD2F0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Linear};

  \node (n26_ggail) at (16.25,-2.8) [draw=c1E293B, line width=0.025cm, fill=cCDEDD0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Softmax};

  \node (n27_76g5v) at (16.25,-0.875) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.5cm, minimum height=1.25cm] {Output Probabilities};

38 \begin{scope}[on background layer]

    \node (n10_ma9el) at (6.25,-13.575) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=7.65cm] {};

    \node (n23_egiii) at (16.25,-11.7) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=11.4cm] {};

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-21.25) -- (6.25,-20.75) -- (6.25,-21.15) -- (6.25,-20.65);

42 \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-19.5) -- (6.25,-19) -- (6.25,-19.1) -- (6.25,-18.6);

43 \draw[->, line width=0.05cm] (n3_x9vjf.east) to[out=0, in=180] (n4_aa0b6.west);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-17.75) -- (6.25,-17.25) -- (6.25,-17.4) -- (6.25,-16.9);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-15.75) -- (6.25,-15.25) -- (6.25,-15.65) -- (6.25,-15.15);

46 \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-14) -- (6.25,-13.5) -- (6.25,-13.65) -- (6.25,-13.15);

47 \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-12) -- (6.25,-11.5) -- (6.25,-11.9) -- (6.25,-11.4);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-21.25) -- (16.25,-20.75) -- (16.25,-21.15) -- (16.25,-20.65);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-19.5) -- (16.25,-19) -- (16.25,-19.1) -- (16.25,-18.6);

50 \draw[->, line width=0.05cm] (n15_te7fn.west) to[out=180, in=0] (n14_e6m10.east);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-17.75) -- (16.25,-17.25) -- (16.25,-17.4) -- (16.25,-16.9);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-15.75) -- (16.25,-15.25) -- (16.25,-15.65) -- (16.25,-15.15);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-14) -- (16.25,-13.5) -- (16.25,-13.65) -- (16.25,-13.15);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-12) -- (16.25,-11.5) -- (16.25,-11.9) -- (16.25,-11.4);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-10.25) -- (16.25,-9.75) -- (16.25,-9.9) -- (16.25,-9.4);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-8.25) -- (16.25,-7.75) -- (16.25,-8.15) -- (16.25,-7.65);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-6.5) -- (16.25,-6) -- (16.25,-5.925) -- (16.25,-5.85) -- (16.25,-5.35);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-4.25) -- (16.25,-3.75) -- (16.25,-3.85) -- (16.25,-3.35);

59 \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-2.25) -- (16.25,-1.75) -- (16.25,-2) -- (16.25,-1.5);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (8.375,-10.825) -- (8.875,-10.825) -- (11.25,-10.825) -- (11.25,-12.575) -- (13.625,-12.575) -- (14.125,-12.575);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (5.825,-18.175) -- (5.325,-18.175) -- (3.625,-18.175) -- (3.625,-14.575) -- (4.125,-14.575);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (4.125,-14.575) -- (4.1,-14.575) -- (3.625,-14.575) -- (3.625,-10.825) -- (4.125,-10.825);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.675,-18.175) -- (17.175,-18.175) -- (18.875,-18.175) -- (18.875,-14.575) -- (18.375,-14.575);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-14.575) -- (18.875,-14.575) -- (18.875,-10.825) -- (18.4,-10.825) -- (18.375,-10.825);

    \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-10.825) -- (18.875,-10.825) -- (18.875,-7.075) -- (18.4,-7.075) -- (18.375,-7.075);

66 \end{scope}

67\end{tikzpicture}

□▭○◯◇▱TΣ

Drag Shapes to Move · Drag Handles to Connect · Space + Drag to Pan · Double-Click to Edit Text

＋−⌗

Output Probabilities

Softmax

Linear

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

N×

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

Add & Norm

Masked Multi-Head
Attention

N×

Positional Encoding

Input Embedding

Inputs

Positional Encoding

Output Embedding

Outputs (shifted right)

▤ Logs　✓ 1.4s⌗　−　131%⌄　＋　◐　↻　

Output Probabilities

Softmax

Linear

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

N×

Add & Norm

Feed Forward

Add & Norm

Multi-Head Attention

Add & Norm

Masked Multi-Head
Attention

N×

Positional Encoding

Input Embedding

Inputs

Positional Encoding

Output Embedding

Outputs (shifted right)

✓Saved to figures/diagram.png

Save As:

diagram.png

Tags:

Where:📁 Desktop　⌃

Source Tree

.gitignore

main.tex

project.json

README.md

references.bib

refs.bib

Structure 38

Filter project map…

main.texmain.tex:19:1

Introductionmain.tex:19:1

System designmain.tex:26:1

fig:architecturemain.tex:43:8

System designmain.tex:26:1

Deployment and evaluationmain.tex:45:1

tab:resultsmain.tex:56:69

Deployment and evaluationmain.tex:45:1

equationmain.tex:50:1

Resultsmain.tex:65:1

figuremain.tex:71:1

Ablations, safety, and limitat…main.tex:81:1

Conclusionmain.tex:87:1

fairmain.tex:92:1

tinymlmain.tex:93:1

References 4

References · 1213Citations · 8Symbols · 4

Filter…

Unresolved citations 4

hastie2009main.tex:21

vaswani2017main.tex:28

Bibliography 1213

references.bib

lecun2015deep2015

Deep learning — Nature 521, 436–444

wilkinson20162016

The FAIR Guiding Principles for scientific…

warden20192019

TinyML: Machine Learning with Ultra-Low…

Citations in main.tex 8

hastie200921 · 67 · 89

wilkinson201623 · 46

vaswani201728 · 85

gelman201329 · 67

lundberg201748 · 83

Labels 4

fig:architecturemain.tex:43

tab:resultsmain.tex:56

equationmain.tex:50

figuremain.tex:71

main.tex

1\documentclass[conference]{IEEEtran}

2\usepackage{amsmath,amssymb,booktabs,graphicx,xcolor,tikz}

3\usepackage[hidelinks]{hyperref}

4\usetikzlibrary{positioning,arrows.meta,shapes.geometric}

5\title{EdgeSense: Energy-Aware Wildfire Detection with Collaborative Sensor Networks}

6\author{\IEEEauthorblockN{Nadia Flores, Min-jun Lee, and Caleb Morgan}

7\IEEEauthorblockA{Center for Resilient Computing, Cascadia Institute of Technology\\

8Seattle, WA, USA \quad \{nflores,mjlee,cmorgan\}@example.edu}}

9\begin{document}\maketitle

10\begin{abstract}

11We present EdgeSense, a low-power wildfire early-warning system that fuses thermal,

12particulate, humidity, and wind observations across a mesh of battery-powered

13nodes. A gated temporal model runs locally and transmits only uncertain windows~\cite{lecun2015deep}.

14In a 16-week controlled field deployment, EdgeSense detected 94.1\% of burn events

15at 0.7 false alarms per node-week while reducing radio energy by 63\% relative to

16periodic streaming. Results use a synthetic but realistic demo dataset.

17\end{abstract}

18\begin{IEEEkeywords}edge computing, environmental sensing, wildfire, TinyML, sensor networks\end{IEEEkeywords}

19\section{Introduction}

20Wildfire ignition can develop faster than sparse satellite revisit cycles and

21manual reports~\cite{hastie2009}. Dense ground networks provide local measurements but create a

22difficult systems trade-off: radios consume most of a node's energy, yet delaying

23communication can hide the earliest evidence~\cite{wilkinson2016}. EdgeSense moves triage to the node

24and treats transmission as a decision under uncertainty.

26\section{System design}

27Each node samples temperature, relative humidity, particulate matter, and wind

28once per minute~\cite{vaswani2017}. A 12\,kB quantized temporal convolutional network scores

29five-minute windows~\cite{gelman2013}. Scores below 0.25 are stored locally; scores between 0.25 and

300.70 trigger peer confirmation; scores above 0.70 trigger an immediate gateway

31packet.

32\begin{figure}[t]\centering

33\begin{tikzpicture}[font=\scriptsize,node distance=4mm,

34n/.style={draw,rounded corners,fill=blue!6,minimum width=18mm,minimum height=7mm},

35a/.style={-{Stealth[length=1.7mm]},thick}]

36\node[n] (s) {Sensors}; \node[n,right=of s] (f) {Feature buffer};

37\node[n,right=of f] (m) {Tiny model}; \node[n,below=of m] (p) {Peer vote};

38\node[n,left=of p] (g) {Gateway};

39\draw[a](s)--(f);\draw[a](f)--(m);\draw[a](m)--(p);\draw[a](p)--(g);

40\draw[a,bend left=25](m) to node[above]{high confidence}(g);

41\end{tikzpicture}

42\caption{EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.}

43\label{fig:architecture}\end{figure}

45\section{Deployment and evaluation}

46We deployed 48 nodes across three ecological zones~\cite{wilkinson2016}. Controlled burns and heated

47aerosol challenges generated 186 positive windows; 11,420 background windows

48captured fog, dust, vehicle exhaust, and sensor maintenance~\cite{lundberg2017}. Splits were grouped

49by day to prevent temporal leakage.

50\begin{equation}

51\mathcal{L}=-w_+y\log p-w_-(1-y)\log(1-p)+\lambda\lVert\theta\rVert_2^2.

52\end{equation}

53We compare periodic streaming, fixed thresholds, an ungated neural baseline, and

54EdgeSense. The primary metrics are event recall, false alarms per node-week,

55median alert latency, and measured radio energy.

56\begin{table}[t]\centering\caption{Field evaluation results.}\label{tab:results}

57\begin{tabular}{@{}lrrrr@{}}\toprule

58Method & Recall & FA/wk & Latency & Energy\\\midrule

59Periodic stream & .957 & 1.8 & 72 s & 100\%\\

60Fixed threshold & .876 & 3.4 & 61 s & 41\%\\

61Ungated TCN & .946 & 1.1 & 58 s & 52\%\\

62\textbf{EdgeSense} & \textbf{.941} & \textbf{0.7} & \textbf{54 s} & \textbf{37\%}\\\bottomrule

63\end{tabular}\end{table}

65\section{Results}

66EdgeSense retains 98.3\% of the streaming baseline's recall while using 37\% of

67its radio energy~\cite{gelman2013}. Peer voting removes 31\% of fog-related false alarms~\cite{hastie2009}. At the

68median observed solar input, the modeled service interval rises from 11 to 29

69months. Performance degrades gracefully when one sensing channel is unavailable.

71\begin{figure}[t]\centering

72\begin{tikzpicture}[x=.72cm,y=.55cm,font=\scriptsize]

73\draw[->](0,0)--(7,0)node[right]{week};\draw[->](0,0)--(0,5)node[above]{recall};

74\draw[blue,very thick] plot coordinates{(0,.8)(1,1.6)(2,2.2)(3,2.8)(4,3.5)(5,4.0)(6,4.25)};

75\draw[orange,very thick,dashed] plot coordinates{(0,.7)(1,1.3)(2,1.9)(3,2.4)(4,2.8)(5,3.1)(6,3.25)};

76\node[blue] at (5.2,4.5){EdgeSense};\node[orange] at (5,2.6){threshold};

77\end{tikzpicture}

78\caption{Cumulative detection performance remained stable across the deployment.}

79\end{figure}

81\section{Ablations, safety, and limitations}

82Removing peer voting increases false alarms by 44\%; removing wind direction

83reduces recall by 3.8 points~\cite{lundberg2017}. EdgeSense is advisory: alerts require confirmation

84by the incident-management system, and nodes never autonomously dispatch a public

85warning~\cite{vaswani2017}. The deployment did not cover crown fires or winter conditions.

87\section{Conclusion}

88Selective communication can extend unattended sensing without hiding early

89signals~\cite{hastie2009}. The next deployment will test transfer across biomes and signed

90over-the-air model updates.

91\begin{thebibliography}{2}

92\bibitem{fair} M~\cite{wilkinson2016}. Wilkinson et al., \emph{The FAIR Guiding Principles}, \emph{Scientific Data}, 2016.

93\bibitem{tinyml} P. Warden and D. Situnayake, \emph{TinyML}. O'Reilly, 2019.

94\end{thebibliography}

95\end{document}

Ask AI

Key, author, or title…

ba2016Ba, Jimmy Lei and Kiros, Jamie Ryan and Hinton, Geoffrey E. · 2016…

bahdanau2015Bahdanau, Dzmitry and Cho, Kyunghyun and Bengio, Yoshua · 2015…

cheng2016Cheng, Jianpeng and Dong, Li and Lapata, Mirella · 2016 · Long S…

cho2014Cho, Kyunghyun and van Merriënboer, Bart and Gulcehre, Cagl…

gehring2017Gehring, Jonas and Auli, Michael and Grangier, David and Yarats…

hochreiter1997Hochreiter, Sepp and Schmidhuber, Jürgen · 1997 · Neural Com…

Find and add a new citation…

Add citation

DOI, arXiv id, URL, or a paper title…10.1038/nature14539Look up

Only the identifier or title is sent, to doi.org, arXiv, or Crossref.

Paste a DOI, an arXiv id, or a URL to fetch the entry directly, or type a title to search Crossref. Try one:

DOI: 10.1038/nature14539 arXiv: 1706.03762 Title: Attention is all you need

@article{lecun2015deep,
  author  = {LeCun, Yann and Bengio, Yoshua and Hinton, Geoffrey},
  title   = {Deep learning},
  journal = {Nature},
  volume  = 521,
  pages   = {436--444},
  year    = 2015,
  doi     = {10.1038/nature14539},
}

CancelAdd to .bib and cite

Added \cite{lecun2015deep} · references.bib

main.texmain.tex (Working Tree)

Working ⇆ Index

| 8 unchanged lines |

9\begin{document}\maketitle

10\begin{abstract}

11We present EdgeSense, a low-power wildfire early-warning system that fuses thermal,

12particulate, humidity, and wind observations across a mesh of battery-powered

13nodes. A gated temporal model runs locally and transmits only uncertain windows.

14In a 16-week controlled field deployment, EdgeSense detected 94.1\% of burn events

15at 0.7 false alarms per node-week while reducing radio energy by 63\% relative to

16periodic streaming. Results use a synthetic but realistic demo dataset.

17\end{abstract}

| 78 unchanged lines |

| 8 unchanged lines |

9\begin{document}\maketitle

10\begin{abstract}

11We present EdgeSense, a low-power wildfire early-warning system that fuses thermal,

12particulate, humidity, and wind observations across a mesh of battery-powered

13nodes. A gated temporal model runs locally and transmits only uncertain windows~\cite{lecun2015deep}.

14In a 16-week controlled field deployment, EdgeSense detected 94.1\% of burn events

15at 0.7 false alarms per node-week while reducing radio energy by 63\% relative to

16periodic streaming. Results use a synthetic but realistic demo dataset.

17\end{abstract}

| 78 unchanged lines |

1\documentclass[conference]{IEEEtran}

2\usepackage{amsmath,amssymb,booktabs,graphicx,xcolor,tikz}

3\usepackage[hidelinks]{hyperref}

4\usetikzlibrary{positioning,arrows.meta,shapes.geometric}

5\title{EdgeSense: Energy-Aware Wildfire Detection with Collaborative Sensor Networks}

6\author{\IEEEauthorblockN{Nadia Flores, Min-jun Lee, and Caleb Morgan}

7\IEEEauthorblockA{Center for Resilient Computing, Cascadia Institute of Technology\\

8Seattle, WA, USA \quad \{nflores,mjlee,cmorgan\}@example.edu}}

9\begin{document}\maketitle

10\begin{abstract}

11We present EdgeSense, a low-power wildfire early-warning system that fuses thermal,

12particulate, humidity, and wind observations across a mesh of battery-powered

13nodes. A gated temporal model runs locally and transmits only uncertain windows~\cite{lecun2015deep}.

14In a 16-week controlled field deployment, EdgeSense detected 94.1\% of burn events

15at 0.7 false alarms per node-week while reducing radio energy by 63\% relative to

16periodic streaming. Results use a synthetic but realistic demo dataset.

17\end{abstract}

18\begin{IEEEkeywords}edge computing, environmental sensing, wildfire, TinyML, sensor networks\end{IEEEkeywords}

19\section{Introduction}

20Wildfire ignition can develop faster than sparse satellite revisit cycles and

21manual reports~\cite{hastie2009}. Dense ground networks provide local measurements but create a

22difficult systems trade-off: radios consume most of a node's energy, yet delaying

23communication can hide the earliest evidence~\cite{wilkinson2016}. EdgeSense moves triage to the node

24and treats transmission as a decision under uncertainty.

26\section{System design}

27Each node samples temperature, relative humidity, particulate matter, and wind

28once per minute~\cite{vaswani2017}. A 12\,kB quantized temporal convolutional network scores

29five-minute windows~\cite{gelman2013}. Scores below 0.25 are stored locally; scores between 0.25 and

300.70 trigger peer confirmation; scores above 0.70 trigger an immediate gateway

31packet.

32\begin{figure}[t]\centering

33\begin{tikzpicture}[font=\scriptsize,node distance=4mm,

34n/.style={draw,rounded corners,fill=blue!6,minimum width=18mm,minimum height=7mm},

35a/.style={-{Stealth[length=1.7mm]},thick}]

36\node[n] (s) {Sensors}; \node[n,right=of s] (f) {Feature buffer};

37\node[n,right=of f] (m) {Tiny model}; \node[n,below=of m] (p) {Peer vote};

38\node[n,left=of p] (g) {Gateway};

39\draw[a](s)--(f);\draw[a](f)--(m);\draw[a](m)--(p);\draw[a](p)--(g);

40\draw[a,bend left=25](m) to node[above]{high confidence}(g);

41\end{tikzpicture}

42\caption{EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.}

43\label{fig:architecture}\end{figure}

45\section{Deployment and evaluation}

46We deployed 48 nodes across three ecological zones~\cite{wilkinson2016}. Controlled burns and heated

47aerosol challenges generated 186 positive windows; 11,420 background windows

48captured fog, dust, vehicle exhaust, and sensor maintenance~\cite{lundberg2017}. Splits were grouped

49by day to prevent temporal leakage.

50\begin{equation}

51\mathcal{L}=-w_+y\log p-w_-(1-y)\log(1-p)+\lambda\lVert\theta\rVert_2^2.

52\end{equation}

53We compare periodic streaming, fixed thresholds, an ungated neural baseline, and

54EdgeSense. The primary metrics are event recall, false alarms per node-week,

55median alert latency, and measured radio energy.

56\begin{table}[t]\centering\caption{Field evaluation results.}\label{tab:results}

57\begin{tabular}{@{}lrrrr@{}}\toprule

58Method & Recall & FA/wk & Latency & Energy\\\midrule

59Periodic stream & .957 & 1.8 & 72 s & 100\%\\

60Fixed threshold & .876 & 3.4 & 61 s & 41\%\\

61Ungated TCN & .946 & 1.1 & 58 s & 52\%\\

62\textbf{EdgeSense} & \textbf{.941} & \textbf{0.7} & \textbf{54 s} & \textbf{37\%}\\\bottomrule

63\end{tabular}\end{table}

65\section{Results}

66EdgeSense retains 98.3\% of the streaming baseline's recall while using 37\% of

67its radio energy~\cite{gelman2013}. Peer voting removes 31\% of fog-related false alarms~\cite{hastie2009}. At the

68median observed solar input, the modeled service interval rises from 11 to 29

69months. Performance degrades gracefully when one sensing channel is unavailable.

71\begin{figure}[t]\centering

72\begin{tikzpicture}[x=.72cm,y=.55cm,font=\scriptsize]

73\draw[->](0,0)--(7,0)node[right]{week};\draw[->](0,0)--(0,5)node[above]{recall};

74\draw[blue,very thick] plot coordinates{(0,.8)(1,1.6)(2,2.2)(3,2.8)(4,3.5)(5,4.0)(6,4.25)};

75\draw[orange,very thick,dashed] plot coordinates{(0,.7)(1,1.3)(2,1.9)(3,2.4)(4,2.8)(5,3.1)(6,3.25)};

76\node[blue] at (5.2,4.5){EdgeSense};\node[orange] at (5,2.6){threshold};

77\end{tikzpicture}

78\caption{Cumulative detection performance remained stable across the deployment.}

79\end{figure}

81\section{Ablations, safety, and limitations}

82Removing peer voting increases false alarms by 44\%; removing wind direction

83reduces recall by 3.8 points~\cite{lundberg2017}. EdgeSense is advisory: alerts require confirmation

84by the incident-management system, and nodes never autonomously dispatch a public

85warning~\cite{vaswani2017}. The deployment did not cover crown fires or winter conditions.

87\section{Conclusion}

88Selective communication can extend unattended sensing without hiding early

89signals~\cite{hastie2009}. The next deployment will test transfer across biomes and signed

90over-the-air model updates.

91\begin{thebibliography}{2}

92\bibitem{fair} M~\cite{wilkinson2016}. Wilkinson et al., \emph{The FAIR Guiding Principles}, \emph{Scientific Data}, 2016.

93\bibitem{tinyml} P. Warden and D. Situnayake, \emph{TinyML}. O'Reilly, 2019.

94\end{thebibliography}

95\end{document}

Import 6

↕ MARKDOWNAlpine eDNA Pilot — Lab Notebook

DocumentUpdated Jul 31

↕ TECTONICMapping Pollinator Corridors — Research Poster

DocumentUpdated Jul 29

↕ TECTONICFederated Learning Data Flow

ImageUpdated Jul 29

↕ TECTONICCatalyst Screening Data Report

DocumentUpdated Jul 29

↕ TECTONICMangrove Canopy Drone Survey

DocumentUpdated Jul 28

Alex Chen

Senior Software Engineer

[email protected] · (650) 555-0142 · Mountain View, CA · alexchen.dev · github.com/alexchen

SUMMARY

Senior Software Engineer (L5) with 9+ years designing and operating large-scale distributed systems. Deep experience in backend infrastructure, performance, and reliability.

EXPERIENCE

Senior Software Engineer (L5)Google

Mountain View, CAMar 2020 — Present

Tech lead for a Search serving component handling 2M+ queries/second; cut p99 latency 38%.
Designed a globally replicated feature store on Spanner backing 40+ ML models.
Led migration of a 300-service fleet to a new RPC framework; reduced on-call pages 45%.
Mentored six engineers and ran the team’s design-review program.

Software Engineer (L4)Google

Mountain View, CAJul 2017 — Mar 2020

Built a real-time aggregation pipeline processing 8B events/day.
Cut batch-job cost 30% with sharding and incremental computation.

Software EngineerStripe

San Francisco, CAAug 2015 — Jun 2017

Shipped ledger and reconciliation services processing $60B+/year.
Reduced a reconciliation run from six hours to 25 minutes.

SELECTED PROJECTS

Ratelgithub.com/alexchen/ratel

Go, Raft, gRPC — distributed rate limiter doing 4M+ decisions/sec/node.

tqlgithub.com/alexchen/tql

Rust — typed query builder for Postgres with compile-time-checked SQL.

EDUCATION

M.S. Computer ScienceStanford University

B.S. Computer ScienceUniversity of Illinois Urbana-Champaign

SKILLS

Languages: C++, Go, Rust, Python, Java, SQL
Systems: Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis

↕ TECTONICSleep and Memory Consolidation — Evidence Map

DocumentUpdated Jul 28

↕ TECTONICUrban Heat Transect Field Protocol

DocumentUpdated Jul 27

↕ TECTONICNSF Progress Report — Year 2

DocumentUpdated Jul 27

↕ TYPSTArjun Rao — Computational Linguistics CV

DocumentUpdated Jul 11

↕ TECTONICOrganoid Response Profiles for Therapy Selection

DocumentUpdated Jul 11

↕ TECTONICSTAT 214 — Causal Inference Problem Set

DocumentUpdated Mar 3

↕ TECTONICDextrous Manipulation — Proposal

DocumentUpdated Jul 9

↕ TECTONICCivic Shade and Pedestrian Heat Exposure

DocumentUpdated Jul 8

↕ TECTONICCS 271 — Approximation Algorithms PS3

DocumentUpdated Jul 5

↕ TECTONICWeak-Lensing Mass Reconstruction of NCJ 1427

DocumentUpdated Jul 2

↕ TECTONICStatistical Consulting Report

DocumentUpdated Jun 25

OleaflyAlex Chen - Google Resume

prajwal-svm

main.tex

T⌄

1\documentclass[11pt,letterpaper]{article}

2\usepackage[T1]{fontenc}

3\usepackage[margin=0.5in]{geometry}

4\usepackage{titlesec}

5\usepackage{enumitem}

6\usepackage{hyperref}

8% ATS-friendly: single column, real selectable text, no tables-for-layout,

9% embedded subset fonts. Linear reading order so parsers extract cleanly.

10\setlength{\parindent}{0pt}

11\pagenumbering{gobble}

12\hypersetup{colorlinks=true, urlcolor=black, linkcolor=black}

14\titleformat{\section}{\large\bfseries\uppercase}{}{0em}{}[\vspace{2pt}\titlerule]

15\titlespacing*{\section}{0pt}{8pt}{4pt}

16\setlist[itemize]{leftmargin=14pt, itemsep=1pt, topsep=2pt, parsep=0pt}

18% role{Title}{Company}{Location}{Dates}

19\newcommand{\role}[4]{%

20 \textbf{#1} \hfill #2 \\

21 \textit{#3} \hfill \textit{#4}%

22}

24\begin{document}

26\begin{center}

27 {\Huge \textbf{Alex Chen}}\\[3pt]

28 \small

29 Senior Software Engineer\\[2pt]

30 \href{mailto:[email protected]}{[email protected]}\,$\cdot$\,

31 (650) 555-0142\,$\cdot$\,

32 Mountain View, CA\,$\cdot$\,

33 \href{https://alexchen.dev}{alexchen.dev}\,$\cdot$\,

34 \href{https://github.com/alexchen}{github.com/alexchen}\,$\cdot$\,

35 \href{https://linkedin.com/in/alexchen}{linkedin.com/in/alexchen}

36\end{center}

37\vspace{2pt}

39\section*{Summary}

40Senior Software Engineer (L5) with 9+ years designing and operating large-scale

41distributed systems. Deep experience in backend infrastructure, performance, and

42reliability, with a track record of shipping products used by hundreds of

43millions of users and mentoring engineers toward senior roles.

45\section*{Experience}

46\role{Senior Software Engineer (L5)}{Google}{Mountain View, CA}{Mar 2020 -- Present}

47\begin{itemize}

  \item Tech lead for a Search serving component handling 2M+ queries/second; drove a redesign that cut p99 latency 38\% and saved an estimated \$14M/year in compute.

  \item Designed and launched a globally-replicated feature store on Spanner backing 40+ ML models, adopted by 12 teams across Search and Ads.

  \item Led migration of a 300-service fleet to a new RPC framework, improving tail latency and reducing on-call pages by 45\%.

51 \item Mentored 6 engineers (2 promoted to senior); ran the team's design-review and readability programs.

52\end{itemize}

53\vspace{3pt}

54\role{Software Engineer (L4)}{Google}{Mountain View, CA}{Jul 2017 -- Mar 2020}

55\begin{itemize}

  \item Built a real-time aggregation pipeline (C++, Flume) processing 8B events/day for a Search-quality dashboard used org-wide.

57 \item Cut batch-job cost 30\% by reworking sharding and introducing incremental recomputation.

58\end{itemize}

59\vspace{3pt}

60\role{Software Engineer}{Stripe}{San Francisco, CA}{Aug 2015 -- Jun 2017}

61\begin{itemize}

62 \item Shipped core ledger and reconciliation services for a payments platform processing \$60B+/year.

63 \item Reduced a reconciliation run from 6 hours to 25 minutes via parallelization and a columnar store.

64\end{itemize}

66\section*{Selected Projects}

67\textbf{Ratel} \hfill \href{https://github.com/alexchen/ratel}{github.com/alexchen/ratel} \\

\textit{Go, Raft, gRPC} --- open-source distributed rate limiter doing 4M+ decisions/sec/node; 1.5k GitHub stars.\par\vspace{4pt}

69\textbf{tql} \hfill \href{https://github.com/alexchen/tql}{github.com/alexchen/tql} \\

70\textit{Rust} --- a typed query builder for Postgres with compile-time-checked SQL.

72\section*{Education}

73\textbf{M.S. Computer Science} \hfill Stanford University \\

74\textit{2013 -- 2015} \hfill \textit{Focus: Distributed Systems}\par\vspace{4pt}

75\textbf{B.S. Computer Science} \hfill University of Illinois Urbana-Champaign \\

76\textit{2009 -- 2013} \hfill \textit{GPA: 3.9 / 4.0}

78\section*{Skills}

79\textbf{Languages:} C++, Go, Rust, Python, Java, SQL \\

80\textbf{Systems:} Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis \\

81\textbf{Focus:} Distributed systems, performance, reliability (SLOs), system design, mentoring

83\end{document}

Logs 1.1s100%⌃　1　of 1　⌄

Compiling1/3 done

1 running · 1 queued

Language analysisCurrent project revision analyzed

DONE

CompilingProducing and verifying PDF output

RUNNING

Rendering PDFWaiting for verified compile output

QUEUED

Alex Chen

Senior Software Engineer

[email protected] · (650) 555-0142 · Mountain View, CA · alexchen.dev · github.com/alexchen

SUMMARY

Senior Software Engineer (L5) with 9+ years designing and operating large-scale distributed systems. Deep experience in backend infrastructure, performance, and reliability.

EXPERIENCE

Senior Software Engineer (L5)Google

Mountain View, CAMar 2020 — Present

Tech lead for a Search serving component handling 2M+ queries/second; cut p99 latency 38%.
Designed a globally replicated feature store on Spanner backing 40+ ML models.
Led migration of a 300-service fleet to a new RPC framework; reduced on-call pages 45%.
Mentored six engineers and ran the team’s design-review program.

Software Engineer (L4)Google

Mountain View, CAJul 2017 — Mar 2020

Built a real-time aggregation pipeline processing 8B events/day.
Cut batch-job cost 30% with sharding and incremental computation.

Software EngineerStripe

San Francisco, CAAug 2015 — Jun 2017

Shipped ledger and reconciliation services processing $60B+/year.
Reduced a reconciliation run from six hours to 25 minutes.

SELECTED PROJECTS

Ratelgithub.com/alexchen/ratel

Go, Raft, gRPC — distributed rate limiter doing 4M+ decisions/sec/node.

tqlgithub.com/alexchen/tql

Rust — typed query builder for Postgres with compile-time-checked SQL.

EDUCATION

M.S. Computer ScienceStanford University

B.S. Computer ScienceUniversity of Illinois Urbana-Champaign

SKILLS

Languages: C++, Go, Rust, Python, Java, SQL
Systems: Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis

Preflight

ATS readinessRun

AccessibilityRe-run

For research, government, and published PDFs

Checks screen-reader readiness against Section 508 / WCAG: missing alt text, document language, reading order, and whether the PDF is tagged.

82Access

Compiled output

PDF is missing a language and titleACCESSIBILITY

Not Section 508 / PDF-UA ready: this PDF is not taggedACCESSIBILITY

References & assetsRe-run

For research and multi-file documents

Finds undefined citations and cross-references, duplicate labels, and missing figures before they break your PDF at submission.

72Refs

3 citations have no bibliography entryREFERENCES · main.tex

main.tex

CodeVisual

Results

EdgeSense retains 98.3% of the streaming baseline’s recall while using 37% of its radio energy gelman2014. Peer voting removes 31% of fog-related false alarms hastie2009. At the median observed solar input, the modeled service interval rises from 11 to 29 months. Performance degrades gracefully when one sensing channel is unavailable.

Ablations, safety, and limitations

Removing peer voting increases false alarms by 44%; removing wind direction reduces recall by 3.8 points lundberg2017. EdgeSense is advisory: alerts require confirmation by the incident-management system.

1\documentclass[conference]{IEEEtran}

2\usepackage{amsmath,amssymb,booktabs,graphicx,xcolor,tikz}

3\usepackage[hidelinks]{hyperref}

4\usetikzlibrary{positioning,arrows.meta,shapes.geometric}

5\title{EdgeSense: Energy-Aware Wildfire Detection with Collaborative Sensor Networks}

6\author{\IEEEauthorblockN{Nadia Flores, Min-jun Lee, and Caleb Morgan}

7\IEEEauthorblockA{Center for Resilient Computing, Cascadia Institute of Technology\\

8Seattle, WA, USA \quad \{nflores,mjlee,cmorgan\}@example.edu}}

9\begin{document}\maketitle

10\begin{abstract}

11We present EdgeSense, a low-power wildfire early-warning system that fuses thermal,

12particulate, humidity, and wind observations across a mesh of battery-powered

13nodes. A gated temporal model runs locally and transmits only uncertain windows~\cite{lecun2015deep}.

14In a 16-week controlled field deployment, EdgeSense detected 94.1\% of burn events

15at 0.7 false alarms per node-week while reducing radio energy by 63\% relative to

16periodic streaming. Results use a synthetic but realistic demo dataset.

17\end{abstract}

18\begin{IEEEkeywords}edge computing, environmental sensing, wildfire, TinyML, sensor networks\end{IEEEkeywords}

19\section{Introduction}

20Wildfire ignition can develop faster than sparse satellite revisit cycles and

21manual reports~\cite{hastie2009}. Dense ground networks provide local measurements but create a

22difficult systems trade-off: radios consume most of a node's energy, yet delaying

23communication can hide the earliest evidence~\cite{wilkinson2016}. EdgeSense moves triage to the node

24and treats transmission as a decision under uncertainty.

26\section{System design}

27Each node samples temperature, relative humidity, particulate matter, and wind

28once per minute~\cite{vaswani2017}. A 12\,kB quantized temporal convolutional network scores

29five-minute windows~\cite{gelman2013}. Scores below 0.25 are stored locally; scores between 0.25 and

300.70 trigger peer confirmation; scores above 0.70 trigger an immediate gateway

31packet.

32\begin{figure}[t]\centering

33\begin{tikzpicture}[font=\scriptsize,node distance=4mm,

34n/.style={draw,rounded corners,fill=blue!6,minimum width=18mm,minimum height=7mm},

35a/.style={-{Stealth[length=1.7mm]},thick}]

36\node[n] (s) {Sensors}; \node[n,right=of s] (f) {Feature buffer};

37\node[n,right=of f] (m) {Tiny model}; \node[n,below=of m] (p) {Peer vote};

38\node[n,left=of p] (g) {Gateway};

39\draw[a](s)--(f);\draw[a](f)--(m);\draw[a](m)--(p);\draw[a](p)--(g);

40\draw[a,bend left=25](m) to node[above]{high confidence}(g);

41\end{tikzpicture}

42\caption{EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.}

43\label{fig:architecture}\end{figure}

45\section{Deployment and evaluation}

46We deployed 48 nodes across three ecological zones~\cite{wilkinson2016}. Controlled burns and heated

47aerosol challenges generated 186 positive windows; 11,420 background windows

48captured fog, dust, vehicle exhaust, and sensor maintenance~\cite{lundberg2017}. Splits were grouped

49by day to prevent temporal leakage.

50\begin{equation}

51\mathcal{L}=-w_+y\log p-w_-(1-y)\log(1-p)+\lambda\lVert\theta\rVert_2^2.

52\end{equation}

53We compare periodic streaming, fixed thresholds, an ungated neural baseline, and

54EdgeSense. The primary metrics are event recall, false alarms per node-week,

55median alert latency, and measured radio energy.

56\begin{table}[t]\centering\caption{Field evaluation results.}\label{tab:results}

57\begin{tabular}{@{}lrrrr@{}}\toprule

58Method & Recall & FA/wk & Latency & Energy\\\midrule

59Periodic stream & .957 & 1.8 & 72 s & 100\%\\

60Fixed threshold & .876 & 3.4 & 61 s & 41\%\\

61Ungated TCN & .946 & 1.1 & 58 s & 52\%\\

62\textbf{EdgeSense} & \textbf{.941} & \textbf{0.7} & \textbf{54 s} & \textbf{37\%}\\\bottomrule

63\end{tabular}\end{table}

65\section{Results}

66EdgeSense retains 98.3\% of the streaming baseline's recall while using 37\% of

67its radio energy~\cite{gelman2013}. Peer voting removes 31\% of fog-related false alarms~\cite{hastie2009}. At the

68median observed solar input, the modeled service interval rises from 11 to 29

69months. Performance degrades gracefully when one sensing channel is unavailable.

71\begin{figure}[t]\centering

72\begin{tikzpicture}[x=.72cm,y=.55cm,font=\scriptsize]

73\draw[->](0,0)--(7,0)node[right]{week};\draw[->](0,0)--(0,5)node[above]{recall};

74\draw[blue,very thick] plot coordinates{(0,.8)(1,1.6)(2,2.2)(3,2.8)(4,3.5)(5,4.0)(6,4.25)};

75\draw[orange,very thick,dashed] plot coordinates{(0,.7)(1,1.3)(2,1.9)(3,2.4)(4,2.8)(5,3.1)(6,3.25)};

76\node[blue] at (5.2,4.5){EdgeSense};\node[orange] at (5,2.6){threshold};

77\end{tikzpicture}

78\caption{Cumulative detection performance remained stable across the deployment.}

79\end{figure}

81\section{Ablations, safety, and limitations}

82Removing peer voting increases false alarms by 44\%; removing wind direction

83reduces recall by 3.8 points~\cite{lundberg2017}. EdgeSense is advisory: alerts require confirmation

84by the incident-management system, and nodes never autonomously dispatch a public

85warning~\cite{vaswani2017}. The deployment did not cover crown fires or winter conditions.

87\section{Conclusion}

88Selective communication can extend unattended sensing without hiding early

89signals~\cite{hastie2009}. The next deployment will test transfer across biomes and signed

90over-the-air model updates.

91\begin{thebibliography}{2}

92\bibitem{fair} M~\cite{wilkinson2016}. Wilkinson et al., \emph{The FAIR Guiding Principles}, \emph{Scientific Data}, 2016.

93\bibitem{tinyml} P. Warden and D. Situnayake, \emph{TinyML}. O'Reilly, 2019.

94\end{thebibliography}

95\end{document}

Logs 1.9s 100% 1 of 2

Compiling 1/3 done

1 running · 1 queued

Language analysisCurrent project revision analyzed
CompilingProducing and verifying PDF output
Rendering PDFWaiting for verified compile output

EdgeSense: Energy-Aware Wildfire Detection with
Collaborative Sensor Networks

Nadia Flores, Min-jun Lee, and Caleb Morgan
Center for Resilient Computing, Cascadia Institute of Technology
Seattle, WA, USA {nflores,mjlee,cmorgan}@example.edu

Fig. 1. EdgeSense performs local triage and requests peer confirmation only for ambiguous windows.

Index Terms—edge computing, environmental sensing, wildfire, TinyML, sensor networks

I. INTRODUCTION

II. SYSTEM DESIGN

III. DEPLOYMENT AND EVALUATION

ℒ = −w₊y log p − w₋(1−y) log(1−p) + λ∥θ∥²₂.(1)

TABLE I
Field evaluation results.

Method	Recall	FA/wk	Latency	Energy
Periodic stream	.957	1.8	72 s	100%
Fixed threshold	.876	3.4	61 s	41%
Ungated TCN	.946	1.1	58 s	52%
EdgeSense	.941	0.7	54 s	37%

Fig. 2. Cumulative detection performance remained stable across the deployment.

IV. RESULTS

V. ABLATIONS, SAFETY, AND LIMITATIONS

VI. CONCLUSION

Selective communication can extend unattended sensing without hiding early signals . The next deployment will test transfer across biomes and signed over-the-air model updates.

REFERENCES

[2]W. Schroeder, P. Oliva, L. Giglio, and I. A. Csiszar, “The new VIIRS 375 m active fire detection data product,” Remote Sensing of Environment, vol. 143, pp. 85–96, 2014.

[6]S. Bai, J. Z. Kolter, and V. Koltun, “An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,” arXiv:1803.01271, 2018.

[8]P. P. Ray, “A review on TinyML: State-of-the-art and prospects,” Journal of King Saud University - Computer and Information Sciences, vol. 34, no. 4, pp. 1595–1623, 2022.

[9]Y. Geifman and R. El-Yaniv, “Selective classification for deep neural networks,” arXiv:1705.08500, 2017.

Zoom in

Zoom out

Fit to width

Fit to height

Reset to 100%

25%

50%

75%

100%

150%

200%

400%

Renamed to "gelman2014" (2 edits in 1 file)

Exported EdgeSense.pdf

A powerful research harness

From first citation to final PDF.

Draft, compile, proofread, search, cite, review, and publish without stitching together separate tools. Oleafly keeps your source, preview, references, AI, and Git history in one workspace you control.

Import what you have

Reconstruct many text-based PDFs into editable LaTeX locally, convert DOCX through a managed Pandoc, or let a vision model transcribe a photo of a page. Optional AI refinement after.

Export where you're going

Export PDF, DOCX, HTML, Markdown, and more from contextual menus. You can also download the whole project as a source ZIP.

Make it yours

Vim mode, light and dark themes, accent colors, seven layout presets, and editable keyboard shortcuts. An editor should fit your hands.

Write in Code or Visual

Flip any LaTeX or Markdown document into a visual editor: headings, lists, bold, links, edited directly, with coordinated undo across both modes. Unsupported commands stay visible as raw blocks instead of being silently dropped, so the round-trip back to source is non-destructive.

SyncTeX, word-level

⌘-click a word in the PDF and land on that exact word in the source. Jump back with ⌘⇧J.

Code intelligence

Go to definition, find references, and rename a label, citation key, or macro across every file.

Slash commands

Type / for a Notion-style insert menu: /figure, /table, /section, /cite, /math.

Citations by DOI

Paste a DOI, arXiv id, or URL. Oleafly fetches the BibTeX, dedupes it, and inserts the \cite.

Bring your whole library

Batch-import BibTeX, RIS, EndNote XML, and Zotero RDF. Everything is deduped against your project before a single entry lands in the .bib.

Command palette

⌘K fuzzy-searches every action in the app: projects, files, settings, the lot.

Model-agnostic AI

Use the model that fits the paper. Connect hosted providers, custom OpenAI-compatible endpoints, or local Ollama. Every file change uses the same approval controls, and MCP clients such as Claude Code and Cursor can connect too. You can also leave AI off.

Research on tap

The agent searches alphaXiv and OpenAlex, pulls paper content, and verifies citations against Crossref before they enter your bibliography.

Built-in proofreading

Proofread spelling and grammar locally with Hunspell and Harper. Oleafly ignores LaTeX commands and math so the checks stay focused on your prose.

An agent that remembers

Durable per-project memory notes and a visible plan checklist survive across chats, so the AI picks up where it left off.

From one page to six hundred

The editor and virtualized PDF viewer stay smooth whether it's a one-page resume or a book-length thesis.

A PDF viewer for real reading

Detach the preview to a second monitor, read two-page spreads, invert colors for late nights, or go fullscreen to present.

Diagram Composer

Describe a diagram and the AI writes the TikZ, compiles the figure in isolation, and fixes overlaps by looking at the render. Or draw it yourself: shapes, edges, and an inspector on a real canvas that round-trips with the generated TikZ. Export transparent high-res PNGs, fully offline.

Three engines, one workspace

LaTeX, Typst, and Markdown projects side by side in the same library, with the same Git history, preview, search, and AI workflow.

\begin{document}Hello, world.PDF compiled ✓

The LaTeX editor

Write in LaTeX, Typst, or Markdown with a live PDF preview and word-level SyncTeX. Compilers are included, so there is no TeX setup or build queue.

edit→compile→verify ✓

The AI assistant

Ask the agent to edit your source, compile the project, and read the resulting PDF. Oleafly is model-agnostic, so you can use a hosted provider, local Ollama, your own endpoint, or no AI at all.

doi:10.1145/3576915@article{gatto24} ✓

The reference manager

Import BibTeX, RIS, EndNote, or Zotero with automatic deduplication. Paste a DOI or arXiv ID to create a reference verified against Crossref.

1 page ✓fork ×3 · per application

The resume builder

Start with an ATS-focused resume template, check the extracted text, and fork a tailored copy for each application without losing earlier versions.

+42 −7

The version control

Automatic Git checkpoints preserve every meaningful state. Compare changes, restore an earlier version, and push to GitHub from the same workspace.

PDFDOCXHTMLZIP

The PDF toolkit

Read long PDFs, reconstruct many text-based PDFs into LaTeX, and export PDF, DOCX, HTML, or a source archive for submission.

The research loop

Research, write, verify, publish.

Move from literature search to submission without losing context. Oleafly keeps your sources, references, figures, compiler output, and history together in one inspectable workflow.

1
Research
Search open research sources, turn DOIs and arXiv IDs into BibTeX, and verify citations before they enter your bibliography.
2
Write
Write in LaTeX, Typst, or Markdown with templates, project-wide code intelligence, and an optional AI agent whose edits you approve.
3
Verify
Compile locally, read clear errors, and jump between the source and PDF. Preflight shows what parsers and screen readers will extract.
4
Publish
Export PDF, DOCX, or HTML, package the source for submission, and push to GitHub from the same folder where the work began.

How it compares

The one research workspace built to replace the rest.

Write, compile, proofread, search literature, manage citations, review PDFs, automate edits, and ship your paper from one place. Oleafly brings the entire research stack into one free, open-source app. See how the alternatives fall short below. Docs.

		Oleafly	OpenAI Prism	Overleaf	Typst Web App	Curvenote	LyX	TeXstudio	Texifier	VS Code + LaTeX	Paperpal	Writefull	Resume builders
Ownership and access	Price	Free and open source. No Oleafly subscription	Free with a ChatGPT account	Free tier. Paid plans add features	Free tier. Pro costs extra	Free tier. Paid plans add private projects	Free and open source	Free, GPL	Paid. Free demo	Free. AI has limits or paid plans	Free tier. Prime costs extra	Free tier. Premium costs extra	Varies. Most use a subscription
	No account required	Yes	No	No	No	No	Yes	Yes	Yes	Yes	No	No	Varies
	Works offline	Write, proofread, preview, and compile offline	No, cloud workspace	No, cloud service	No. The web app needs a connection	No. The editor is a cloud service	Yes	Yes	Yes	Yes	No	No	Rarely
	Files stay on your disk	Local folders you own. Self-hosted cloud sync	Cloud storage	Cloud storage	Cloud storage. The compiler can run locally	Cloud storage	Yes	Yes	Local files and file providers	Yes	Cloud service and editor integrations	Files stay in Word or Overleaf	Usually cloud storage
	Version history	Automatic Git checkpoints with diff and restore	Cloud revisions	Full history on paid plans	Git sync on Pro	Built-in version history	Git and other version control	Manual or SVN	Uses file providers or Git tools	Built-in Git that you configure	Not documented	Uses the host app history	Varies
	Real-time collaboration	Coming soon. Unlimited collaboration. Self-hosted	Unlimited collaborators	Built in. Limits depend on the plan	Built in	Real-time editing and comments	No live co-editing	No live co-editing	No live co-editing	Via Live Share	Uses Word, Google Docs, or Overleaf	Uses Word or Overleaf	Varies
	Cloud sync	Coming soon. Browser access while files stay on your machine	Cloud workspace	Cloud workspace plus paid Git and Dropbox sync	Cloud workspace plus Pro Git sync	Cloud projects and connected publishing	Use Git or your own sync service	Use Git or your own sync service	Texifier Connect live sync	Use Git or your own sync service	Cloud workspace	Works inside Word or Overleaf	Usually cloud storage
	Team spaces	Coming soon. Shared templates, libraries, and review flows	Shared projects with unlimited collaborators	Group and institutional plans	Shared projects plus on-premises for organizations	Team workspaces and access controls	No built-in team space	No built-in team space	Shared projects through Texifier Connect	GitHub organizations and Live Share	Institutional dashboards and license management	Institutional access	Varies
Writing and publishing	Document engines	LaTeX, Typst, and Markdown in one workspace	LaTeX only	LaTeX only	Typst only	MyST Markdown and Jupyter	LaTeX and DocBook	LaTeX only	LaTeX and Markdown	Via extensions	Word, Google Docs, web, and Overleaf	Word and Overleaf LaTeX	Resume formats only
	TeX setup	Bundled Tectonic and Typst. No TeX install	No install. Compiles in the cloud	No install. Compiles in the cloud	No TeX install	No TeX install	Install a TeX distribution	Install a TeX distribution	Built-in typesetter and package manager	Install extensions and TeX	Uses Overleaf for LaTeX	Uses Overleaf for LaTeX	No TeX setup
	Code + visual editing	Source plus visual editing with lossless round trips	Source and preview	Source and rich text mode	Source and live preview	Visual editor with LaTeX math	Structured visual editor	Source only	Source and PDF preview	Source only	Visual writing assistant	Inline writing suggestions	Visual only
	Project-wide code intelligence	Autocomplete plus definitions, references, and project-wide rename	Editor and AI assistance	Autocomplete and reference search	Errors, completion, and Assists	Structured references and equations	Labels, references, outline, and completion	Structure and reference checker	Completion, outline, references, and search	LaTeX Workshop IntelliSense	No LaTeX code tools	TeXGPT for LaTeX code	No
	Source and PDF sync	Source and PDF sync at word level, both directions	Instant PDF preview	Source and PDF sync	Live preview	No source and PDF sync	Source and output preview	Almost word-level	Source and PDF sync	Direct and reverse SyncTeX	No	Uses the Overleaf preview	No
	PDF reading modes	Built-in viewer plus detach, spreads, dark, and reader modes	Built-in preview	Built-in preview	Built-in preview	Document preview	Uses a PDF viewer	Integrated viewer	Integrated PDF viewer	Extension viewer with invert mode	Chat with PDF	No PDF reader	Document preview
	Document templates	100+ starters plus reusable custom templates	Not documented	Large publisher template gallery	Typst Universe and private Pro templates	Professional templates on paid plans	Built-in document classes	Built-in and user templates	Not documented	Snippets and community templates	Not documented	No	Core feature
	Project and document import	Files and ZIPs plus PDF, DOCX, Markdown, images, and folders	Folder, ZIP, or drag and drop	Files, ZIP, and paid integrations	Converts LaTeX, Word, and Markdown	Jupyter content and reference imports	LaTeX and other text formats	Open local source files	Local folders and cloud files	Open local source files	PDF and document uploads	Word and Overleaf documents	Usually PDF or DOCX
	Export formats	PDF and source ZIP plus DOCX, HTML, and Markdown	PDF and project files	PDF and source ZIP	PDF, PNG, and SVG	PDF, Word, LaTeX, Typst, and MyST	PDF, HTML, Word, RTF, ePub, and more	PDF and source files	PDF and source files	Depends on local tools	Word workflows	Tracked Word changes	Usually PDF and DOCX
Research and AI	AI model choice	Model agnostic. Hosted providers, custom endpoints, or local Ollama	OpenAI models built in	Built-in research AI. Daily allowance varies by plan	Compiler Assists	Not documented	None	None	None	Copilot, BYOK, or extensions	Paperpal AI	Writefull models and GPT	Built in. Model choice varies
	Local AI	Local Ollama. No hosted provider required	No	No	No	No	No	No built-in AI	No	BYOK or extensions	No	No	No
	Review AI edits before applying	Reviewable diffs plus per-edit approval controls	Keep or Undo	Review AI suggestions	Review each Assist	Not documented	No built-in AI	No built-in AI	No built-in AI	Review changes	Review writing suggestions	Track Changes	Varies
	AI checks the rendered result	Edits source, compiles, then checks the rendered PDF	You recompile to confirm	You recompile to confirm	Compile preview	No	No built-in AI	No built-in AI	No built-in AI	Build checks without PDF review	No rendered PDF review	No rendered PDF review	No
	Literature and citation search	Literature search across alphaXiv and OpenAlex with Crossref verification	Built-in literature search	Reference search plus AI citation suggestions	No built-in search	DOI citation lookup	No built-in search	No	No built-in search	Via AI or extensions	Search across 250M+ articles	Writefull Cite finds missing citations	No
	Autonomous research agents	Coming soon. Agents search literature, verify citations, draft, and hand you the diff	Not documented	Not documented	No	Not documented	No	No	No	Via AI extensions	No	No	No
	Reference library import	BibTeX, RIS, EndNote, and Zotero with pre-import deduplication	Zotero sync	Reference integrations on paid plans	Zotero and Mendeley sync on Pro	Zotero, Mendeley, and EndNote	BibTeX files	Works with BibTeX files	BibTeX files	Via extensions	Citation generator	No library import	No
	MCP for external AI agents	Built-in MCP server for Claude Code, Cursor, and other agents	No	No	No	No	No	No	No	Yes	No	No	No
	Spell and grammar checks	Private offline spelling and grammar with LaTeX-aware prose checks	AI proofreading	Built-in spellcheck plus research-tailored AI feedback	Not documented	Not documented	Hunspell, Aspell, and thesaurus	Interactive spell and grammar checks	Spellcheck	Via extensions or AI	Academic grammar and style	Academic language feedback	Basic checks vary
	Visual and AI diagram tools	Visual canvas plus AI-generated TikZ with render checks	Image and voice to code	Manual TikZ and AI tools	Typst packages and native drawing	Interactive charts and figures	Visual tables and image tools	Manual TikZ with editor helpers	Manual TikZ	Manual or extensions	No	AI tables, equations, and figures	No
Quality and submission	Submission preflight	Compile checks plus venue, ATS, accessibility, references, and privacy	AI review without a submission suite	Compile checks	Compiler errors and Assists	SCMS compliance tools	Compiler logs and optional ChkTeX	Errors and warnings	Clear compiler errors	Linting via extensions	30+ submission checks	Language quality review	Resume checks only
	ATS resume checks	Scored ATS checks plus extracted-text preview	No	No	No	No	No	No	No	No built-in check	No	No	Varies. Scoring is often unclear
	Accessibility preflight	Accessibility checks plus tagged-export preparation	No	No	Not documented	Structured web publishing	Structured documents. No preflight	No	Not documented	No built-in check	No	No	Varies

Privacy, precisely

Local by default. Network by choice.

Your projects stay on your machine. Oleafly uses the network only for features you choose, and shows crash reports before sending them. No Oleafly account. No telemetry. Privacy.

Always local

Project files, drafts, and automatic Git history
Writing, compiling, PDF review, and preflight checks
Offline spellcheck, grammar, diagrams, and optional Ollama

Network, only when you act

Hosted AI receives only the context you choose to send
GitHub sync and research searches run when you request them
Packages, templates, fonts, and app updates download as needed

Philosophy

Your research should outlive the tools used to create it.

Yourthesisshouldn'tneedfivesubscriptions.YourAIshouldn'tguessateditsitneverchecked.Yourreferencesshouldn'ttakeanafternoon.YourfirstPDFshouldn'twaitinabuildqueue.Yourtoolsshouldn'tcostmorethanyourresearch.Onefreeapp.Allofit.

The Oleafly Club: an illustrated celebration of researchers at work, complete with a robot dog, math jokes, and a thesis crown

Open source, seriously built

Independent by design. Built for real work.

Oleafly is built in the open. Reliability is part of the product, from cancellable compiles and safe file paths to Git recovery, PDF fallback, and tests across macOS, Windows, and Linux.

Inspect the code, report an issue, contribute a fix, or star Oleafly on GitHub.

FAQ

What to know before you switch.

What is Oleafly?

Oleafly is a local-first workspace for writing and shipping research. Draft in LaTeX, Typst, or Markdown, compile with bundled engines, proofread spelling and grammar locally, preview PDFs, manage citations, review AI edits, track versions in Git, and check resumes for ATS readability without moving between apps.

Is it really free?

Yes. Oleafly is free and open source. The desktop app does not add a subscription. If you connect a paid AI provider, that provider may charge for usage. Local models through Ollama do not require a paid API.

Which AI models does it work with?

Oleafly is model-agnostic. Use models from hosted providers, run local models through Ollama, or connect a custom OpenAI-compatible endpoint. You choose the model and supply credentials only when the provider requires them.

Do I need an account?

No. The desktop app has no Oleafly account or login. Your projects stay in ordinary local folders. Connected AI and Git hosting services may require their own credentials.

Which tools can it replace?

For many workflows, Oleafly can replace separate writing, proofreading, compilation, citation, PDF, resume, AI, and Git clients. It still works with ordinary files and GitHub, so you can adopt it without locking your work into a proprietary format.

Can I import an existing paper or resume?

Yes. Open an existing folder or import a ZIP, including exports from online LaTeX editors. Oleafly detects the main document. It can convert text-based PDFs to LaTeX locally, route DOCX through Pandoc, and use a vision model to transcribe scans or photos.

Is the PDF output ATS-friendly?

ATS readiness depends on the template and content, so Oleafly does not promise a score. Its resume templates use selectable Unicode text, embedded fonts, single-column layouts, and linear reading order. Preflight shows what a parser actually extracts before you send the PDF.

What about collaboration and sync?

Today, collaboration uses Git. Publish to GitHub, then push and pull from inside Oleafly. Real-time co-editing, comments, and managed sync are planned, but they are not in the app yet.

Built-in proofreading

Proofread spelling and grammar locally with Hunspell and Harper. Oleafly ignores LaTeX commands and math so the checks stay focused on your prose.

Download

Free. Open source. Yours.

Grab a prebuilt installer from the latest release, or build from source. No account, no telemetry.

macOS Windows Linux

Some builds may trigger an operating-system warning on first launch. The install guide explains what to check and how to continue.

Write (()=>{var e=async t=>{await(await t())()};(self.Astro||(self.Astro={})).load=e;window.dispatchEvent(new Event("astro:load"));})();research papersresumesthesesgrant proposalsresearch papers10x faster, all in one free app.

research papers

From first citation to final PDF.

Import what you have

Export where you're going

Make it yours

Write in Code or Visual

SyncTeX, word-level

Code intelligence

Slash commands

Citations by DOI

Bring your whole library

Command palette

Model-agnostic AI

Research on tap

Built-in proofreading

An agent that remembers

From one page to six hundred

A PDF viewer for real reading

Diagram Composer

Three engines, one workspace

The LaTeX editor

The AI assistant

The reference manager

The resume builder

The version control

The PDF toolkit

Research, write, verify, publish.

Research

Write

Verify

Publish

The one research workspace built to replace the rest.

Local by default. Network by choice.

Always local

Network, only when you act

Your research should outlive the tools used to create it.Your research should outlive the tools used to create it.

Independent by design. Built for real work.

What to know before you switch.

Free. Open source. Yours.

Write
research papers
10x faster, all in one free app.

Your research should outlive the tools used to create it.