- Python
- C
- C++
- PyTorch
- deep learning
- scikit-learn
- PySpark
- Databricks
- SQL
- CMake
- gRPC
- PostgreSQL
- applied cryptography
- X25519
- ChaCha20-Poly1305
- pytest
- Catch2
- vitest
- LLM agents
- vLLM
- Ollama
- backtesting
- financial modeling
- econometrics
- Git
- CI/CD
- Docker
[ 01 / 07 ]
About
I like work I can check. Each project page says what I did, what it measured, and what did not work.
A post-quantum mutual-authentication protocol in C11 with fuzz targets, nightly mutation testing and a ProVerif model. A C++23 engine for strategy games with a deterministic lockstep simulation and sandboxed WebAssembly mods. One raycaster that runs on a TI-84 Plus CE calculator and a Garmin watch. A reproduction of a NeurIPS 2025 paper, with a preprocessing fix for its cross-lab failure.
- projects written up on this site
- 36
- commits in the last year, private repos included
- 2,576
- repositories those commits went to
- 72
- best final Kaggle finish
- top 9%
- LeetCode problems solved
- 1,046
Languages
+4 more
Machine learning
+2 more
Data
+3 more
Systems
+2 more
Security
+7 more
MoreTesting, AI and agents, Quant, Tools
Testing
+1 more
AI and agents
Quant
Tools
+1 more
[ 02 / 07 ]
Selected work
- 0.45 mshandshake2026
mldsa-auth
Post-quantum mutual authentication and session protocol in C11, with a login daemon on top.
- in progress
Atlas
C++23 engine for map-based strategy games with a deterministic lockstep simulation and sandboxed WebAssembly mods.
- 2026
LFP2Vec audit
Reproduction of LFP2Vec (NeurIPS 2025) on public Neuropixels data, a preprocessing fix for its cross-lab failure, and two measurements the paper does not make.
- 2026
EEG dementia re-evaluation
Honest, subject-level re-evaluation of a course project that classified Alzheimer's, frontotemporal dementia and healthy controls from EEG: about 60% balanced accuracy, and no deep model beats spectral features.
- 2026
doom-ports
DOOMCE and TRENCHFIRE: one from-scratch raycaster for a TI-84 Plus CE calculator and a Garmin Venu X1 watch.
- 2026
TideFit
Security audit, Next 14 to 16 upgrade and production launch of a roommate's Next.js trip planner.
- 2026
ForgeLog
Offline Android gym tracker with no accounts and no network permission.
- 2026
Custodian
Household inventory that watches CPSC, NHTSA and FDA recalls, served as an MCP server with its own OAuth 2.1 server for Alexa+ and Claude.
- 109/129gold checksin progress
Gemma 4 SWE agent
Solo Kaggle SWE-agent entry, built on a re-implementation of the organizers' unreleased evaluation harness.
- IndexAll 36 projects
[ 03 / 07 ]
Live demos
These run in your browser, on this site. Each one says exactly what it is.
- 01
mldsa-auth
C11 handshake compiled to WebAssembly
- 02
Atlas
Lab app, compiled with Emscripten
- 03
LFP2Vec audit
Interactive charts of saved result files
- 04
EEG dementia re-evaluation
Interactive charts of saved predictions
- 05
doom-ports
Unmodified src/main.c compiled to WebAssembly
- 06
Custodian
TypeScript MCP server over recorded fixtures
- 07
Blockwave
Unchanged Python and NumPy on Pyodide
- 08
gtnh-agent
TypeScript agent loop, mock world
- 09
ephys-mcp
v0.5.1's Python source in Pyodide
- 10
Ledger
Verification code over a recorded run
- 11
NYC AutoDDG
pandas profiler running unchanged in Pyodide
- 12
Hawk signature study
hawk_crypto and pytest in Pyodide
- 13
One-time-pad allocation
protocol.py running unmodified in Pyodide
- 14
Non-stationary bandits
Five algorithms running unmodified in Pyodide
- 15
Snakes and Ladders as a Markov chain
model.py running unmodified in Pyodide
[ 04 / 07 ]
Competitions
01Biohub Cell Tracking, KaggleTop 15%2026
Final rank 591 of 3,947 teams (top 15%); private leaderboard 0.918, public 0.952
The project02Kaggle Playground S6E9Top 9%2026
Out-of-fold AUC 0.9465 from nested target encodings, a LightGBM, CatBoost and MLP blend and 192 experiments judged by paired DeLong tests. Final rank 287 of 3,575 teams (top 9%), private AUC 0.94545
The project03Gemma 4 Developer Agent, Kaggle (Google DeepMind)In progress2026
LLM agent entry in Google ADK; local evaluation harness with Docker sandboxes and tool calls, served by vLLM. In progress, ends Dec 2, 2026
The project04TrafficFlowBenchIn progress2026
Public leaderboard 0.889 against a 0.553 baseline (LightGBM queue forecasts, a PyTorch U-Net, an NNLS origin-destination solver). In progress, ends Nov 7, 2026
The project05Money Talks 2, AI x Fintech Hackathon (Block Convey), Money Operations trackResult pending2026
Solo entry Ledger, a forecast agent that checks every claim against the CRM; 94% recall and 0 false alarms on 235 deals. Result pending
The project[ 05 / 07 ]
By the numbers
updated Oct 25
Shipped
Code
GitHub
@happyc0der →2,001 contributions in the last year
- Public repos
- 34
- Merged PRs
- 35
- Stars
- 8
730 in the last 7 days · 12-day streak
Languages
Share of code across 34 public repos
When I commit
10pm busiest hour
2,576 commits in a year across 72 repos, private ones included · 32% between 10pm and 5am ET · busiest day Thu
Competitive programming
- Codeforces
- 1381 rating
- LeetCode
- 1046 solved
- Hard solved
- 109
- CF contests
- 33
Typing · TypeRacer
lmao_keke →82 WPM over the last 10 races
- Certified
- 90 WPM
- All races
- 82 WPM
- Best race
- 120 WPM
- Races
- 2,444
AI
Claude Code
10.26B tokens last week
43B tokens since Sep 2026, across 2 machines. Most are cached context re-read each turn; the model wrote 83.2M.
Where the tokens went
By model: 67% Opus 5.5 · 21% Opus 5 · 10% Fable 5.1 · 0% Sonnet 5.5 · 1% Other
gtnh-agent · Minecraft
7 quests completed by the agent
- Verified actions
- 2,008
- Blocked by safety
- 291
GregTech: New Horizons on a private server · last ran Oct 5
Health
Garmin · last night
- Sleep
- 6h 07m · score 79
- Resting HR
- 59 bpm
- HRV
- 69 ms · low
- Readiness
- 61 / 100
Steps, last 7 days · full-day average 6.6K a day · 5h 55m sleep a night
Training · 2026
95 workouts, 71 hours
208 km swum this year
- Pool swims80
- Open water4
- Walks4
- Indoor rides3
Last logged Aug 31
Shared to Strava this year: 6 rides, 58.3 km
Personal records
- Longest swim
- 3.8 km
- Swim 1500
- 24:21 min
- Swim 400
- 5:16 min
- Longest ride
- 25.4 km
Pool distances in meters, from Garmin; times that look like watch glitches are left out
Play
Overwatch 2
Master 4 open queue
- Win rate
- 58.8%
- KDA
- 4.29
- Games
- 932
- Played
- 131 h
DankAxon · season 25
Steam
2,740 hours across 378 games
- Cyberpunk 2077205 h
- Europa Universalis IV190 h
- Hearts of Iron IV172 h
- Ghost Recon Breakpoint132 h
Emulators
26 games across 4 emulators
- PSP · PPSSPP · PC + Android19
- PlayStation 3 · RPCS3 · PC4
- Xbox 360 · Xenia · PC2
- PlayStation 2 · AetherSX2 · Android1
Spotify · last 4 weeks
- Blue Imagined#1
- PinkPantheress#2
- GRAZER#3
- atmos bloom#4
- Mohit Chauhan#5
Top track: Sleepwalking · Bring Me The Horizon
[ 06 / 07 ]
Education
New York University, Tandon School of Engineering
Master of Science (MS) in Computer Science
Indraprastha Institute of Information Technology (IIIT) Delhi
Bachelor of Technology (B.Tech) in Computer Science and Applied Mathematics, minor in Economics
Coursework, both degreesData Structures and Algorithms · Algorithms II · Database Systems · Distributed Systems · Operating Systems · Professional Software Development · Machine Learning · Neuroinformatics · Foundations of Data Science · Advanced Machine Learning · Convex Optimization · Big Data · Principles of Database Systems · Information Visualization · Statistical Inference · Applied Cryptography · Post-Quantum Cryptography · Computer Networks · Design and Analysis of Algorithms II · Stochastic Processes · Game Theory · Econometrics · Valuation and Portfolio Management
Resumes