01 · Roasts
The Six-Month Hibernation
Your heatmap is a flat line from October through March — 22 weeks of near-zero commits — then suddenly 1,114 commits like you drank a case of Red Bull. Consistency is a habit, not a sprint.
TypeScript Supremacist
84% TypeScript. You're one language away from just having a config file. Python shows up at 13% like a shy roommate who pays rent but never speaks at dinner.
The Leak Archivist
You made a whole repo — claude-code — to back up Anthropic's leaked sourcemaps before they got DMCA'd. Bold archival energy, but 'I forked leaked code' is a strange line item on a Berkeley bio-ML résumé.
Sprint Engineer, Not Marathon Runner
yc-aixbio-hackathon: 30 commits in 7 days. pretext-image-engine: 23 commits in 13 days. wgs-analysis-pipeline is your only repo with 6+ months of history. You build fast and abandon faster.
Readme-First, Tests-Later (Maybe Never)
9 out of 12 repos have no test suite. Enso-Atlas and wgs-analysis-pipeline are the responsible adults in the room; the other 10 repos are just vibes and READMEs.
Built using
Zoral
Shadows one worker for a week, then takes over their job with zero extra setup. Behaves exactly like the original.
zoral.ai
02 · Category breakdown
- Impact25% weight62C
- Consistency20% weight65C
- Quality20% weight72B
- Depth15% weight65C
- Breadth10% weight55D
- Community10% weight55D
03 · Stats
365-day commit heatmap
143 active days
Language distribution
- TypeScript84%
- Python13%
- Shell1%
- JavaScript1%
- HTML0%
- Jupyter Notebook0%
- Other1%
04 · Numbers
Owned repos
non-fork
19
Commits
last 12 months
1,114
Followers
18
Joined GitHub
Oct 2019
05 · Top repos
Hilo-Hilo /
Enso-Atlas
Enso Atlas is a modular, project-scoped pathology AI platform supporting multiple foundation models and cancer prediction tasks. Built on FastAPI with TransMIL inference, configurable embeddings, and MedGemma report generation. Typed Python backend, documented architecture, comprehensive test suite.
Hilo-Hilo /
pretext-image-engine
TypeScript image-layout library with full test & CI coverage, well-structured multi-file codebase (57KB), comprehensive docs, and npm packaging. Brand-new repo (14 days old) with 23 recent commits showing active development.
Hilo-Hilo /
wgs-analysis-pipeline
Memory-optimized 16GB RAM WGS pipeline with end-to-end workflow (QC, trimming, alignment, variant calling), typed Python sample registry, comprehensive shell scripts with hardware auto-detection, CI/CD via GitHub Actions, and structured docs. Active portfolio project with non-trivial architecture.
Hilo-Hilo /
yc-aixbio-hackathon
Y Combinator AI×Bio hackathon winner: tri-modal LUAD clinical decision support system fusing H&E pathology, RNA transcriptomics, and clinical data. Fast-moving 1-week sprint (30/30 commits, 2075 KB) with functional FastAPI backend, React web UI, and typed Python services. Research-only proxy without clinical license, b
Hilo-Hilo /
stack-stratification
Early-stage LUAD benchmarking framework comparing ARC Stack embeddings to PCA baselines on single-cell transcriptomics. Typed Python with structured layout, meaningful docs (README + ARCHITECTURE.md + STATUS.md), and runnable analysis scripts, but lacks tests, CI, and production maturity.
Hilo-Hilo /
Stack-Benchmarking
Research benchmarking project for Stack foundation model on cancer drug response prediction. Early-stage academic work with clear methodology but lacks tests, CI, and type hints. 25KB codebase with 7 commits in 13 days suggests focused burst development.
Hilo-Hilo /
DeepSeek-OCR-2-Web
DeepSeek OCR 2 Web is a multimodal document parsing system coupling DeepSeek-OCR-2 with Next.js frontend and FastAPI backend, optimized for DGX Spark. Early-stage burst project: ~140KB codebase, single-day commit (Feb 13), untyped Python, no tests/CI, README present but limited documentation. Well-structured services (
Hilo-Hilo /
resume-public
Personal resume distribution repo auto-published by private Actions workflow. Serves static PDF + metadata.json for hansonwen.dev; minimal self-contained scope with 122 KB and 5 commits over 3 days, no tests/CI/license.
Hilo-Hilo /
ptbxl-benchmark
ECG classification benchmark repo with well-documented clinical motivation but minimal codebase (19 KB), no tests, no CI, no implementation visible, and only 4 commits in 15 days—appears to be a specification/template project awaiting agent implementations.
Hilo-Hilo /
claude-code
A leaked sourcemap archive documenting Claude Code's internals with minimal original contribution—essentially a backup/analysis repo of third-party source discovered via npm security oversight.
Hilo-Hilo /
math-distillation-equational-theories-stage1-hive
Hive task scaffold for SAIR math competition prompt optimization; minimal bare-bones setup with single modifiable file (prompt_template.txt), no tests or CI, created and pushed same day.
Hilo-Hilo /
denovo-protein-design
Academic iGEM archive of 2 Jupyter notebooks for RFdiffusion protein design and AutoDock validation. Created and abandoned same day, no tests/CI/license, minimal infrastructure. Scholarly documentation but experimental scope.
06 · Timeline
- Oct 29, 2019Joined GitHub
- Aug 30, 2025Created wgs-analysis-pipeline — Whole genome sequencing analysis pipeline and tools
- Jan 29, 2026Created Enso-Atlas — Universal on-premise pathology AI platform: plug in any foundation model, dataset, or cancer task hassle free.
- Feb 13, 2026Created DeepSeek-OCR-2-Web
- Feb 20, 2026Created denovo-protein-design — A two-step computational pipeline for the de novo design of novel protein binders and their structural validation. The pipeline leverages deep learning-based generative models for
- Feb 26, 2026Created Stack-Benchmarking — Benchmarking the Stack Foundation Model for out-of-distribution cancer drug response prediction with Evo 2 genotype augmentation
- Mar 2, 2026Created yc-aixbio-hackathon — Winning project at Y Combinator Bio/AI Hackathon: Tri-modal LUAD clinical decision support system; H&E pathology + RNA transcriptomics + clinical data fusion for treatment sequenci
- Mar 27, 2026Created ptbxl-benchmark — PTB-XL 12-lead ECG classification benchmark for Hive
- Mar 31, 2026Created claude-code — Claude Code's Source Code & Breakdown from a leaked map file in their NPM registry
- Apr 3, 2026Created stack-stratification — LUAD benchmarking with ARC Institute Stack embeddings
- Apr 6, 2026Created pretext-image-engine — Standalone image-aware text layout engine built on Pretext.
- Apr 18, 2026Created resume-public — Public companion to the private resume repo. Serves website-resume.pdf + metadata.json for hansonwen.dev. Auto-published by Actions; do not hand-edit.
- Apr 19, 2026Created math-distillation-equational-theories-stage1-hive — Hive task for prompt optimization on the SAIR Mathematics Distillation Challenge: Equational Theories Stage 1
- Apr 21, 2026Most recent push to resume-public
07 · Compare
08 · Rubric
How this score was produced
Overall = Σ (category × weight) + gentle top-end curve
Tier thresholds
▸ How the pipeline works
- 01Scrape.Pull every non-fork repo pushed in the last 90 days, plus your contribution calendar, followers, and language byte counts — straight from GitHub's REST & GraphQL APIs.
- 02Triage.A small model reads every repo's file tree + README and picks the 20 files per repo that actually reveal how you code.
- 03Grade each repo. All repos run in parallel through a fast scoring model that reads the picked files and rates each one independently on Impact, Quality, and Depth — with evidence citations.
- 04Aggregate. A larger reasoning model combines the per-repo scores with server-computed stats (heatmap, commit cadence, language entropy, follower count) to produce the 6-dimension profile score + roasts.
- 05Correct.Deterministic server-side checks enforce anchor-scale floors (e.g. a profile with 2,000+ public commits can't score 30 Consistency) and recompute the final verdict.
~90 seconds per profile, ~$0.25 in compute. Total of ~240 files read across your top-12 repos. One rating per GitHub account per day.
▸ Data sources & caveats
- Heatmap & commit totals: GitHub GraphQL
contributionsCollection— covers the last 365 days, includes private repos when the user has opted in (default). - Language %: byte totals across the top 30 owned non-fork repos.
- Curve: a small upward nudge centered on raw score ≈ 70, capping at 100. Prevents specialists from being unfairly penalised for narrow breadth.
- Anchor corrections: when server-measured signals (e.g. privateWorkLikely, multiRepoVolume, follower count) mandate a minimum category score, the aggregation step enforces it. These are signal-conditional, not identity-based floors.