F1 machine learning project.
A recruiter-facing machine learning project with a live Python inference API, a 3D season explorer, and historical evaluation that stays honest about misses.
§ 01 · Featured work
Three projects.
Evaluation harness for coding agents.
A lightweight, reproducible harness for stress-testing AI coding assistants on real tasks, with deterministic runs and review-ready reports.
Reserved for the next project.
Space kept for the next serious piece. I'd rather have three strong projects than ten projects worth skimming.
§ 01 · Featured work
All projectsThree projects,scroll to meet them.
F1machinelearningproject.
A recruiter-facing machine learning project with a live Python inference API, a 3D season explorer, and historical evaluation that stays honest about misses.
LightGBMCalibrationNext.jsFastAPIRailway
Evaluationharnessforcodingagents.
A lightweight, reproducible harness for stress-testing AI coding assistants on real tasks, with deterministic runs and review-ready reports.
EvalsTypeScriptTooling
Write-up soon
eval-harness · run 0147
$ harness run --agent claude-sonnet --suite real-tasks
▸ loaded 142 test cases · 8 capability slices
▸ booting sandboxes · 12 parallel · max 180s
▸ seeded rng · 20260423 · deterministic
✓ 128 / 142 passed · 14 failures · 2 regressions
→ report → ./runs/0147/report.html
Reservedforthenextproject.
Space kept for the next serious piece. I'd rather have three strong projects than ten projects worth skimming.
Write-up soon
★
Fig. — placeholder stampStatus · reserved
§ 02 · About
Theworkshouldexplainitself.
This portfolio is meant to show more than finished visuals. It shows how I frame technical problems, build with AI assistance, and package the result so another person can understand it fast.
I'm most interested in work that sits between product thinking, frontend execution, and technical depth — interfaces, prototypes, tooling, and machine-learning projects that need to be both useful and legible.
AI-native workflow
I use AI coding tools as leverage for speed and iteration — not as a substitute for judgment. I know when to steer, when to stop, and when to rewrite.
Technical storytelling
I turn complex technical work into interfaces and narratives that a non-specialist can still understand quickly. Documentation is part of the product.
Product thinking
I shape the work around what an employer, user, or stakeholder needs to understand first. Craft is the default; shipping is the measure.
§ 03 · How I build
Fast,critical,reproducible.
Prototype quickly
Use AI coding tools to collapse the first draft and get to something testable early. Learning happens inside the loop, not before it.
Own the structure
Keep the architecture, naming, and tradeoffs intentional. AI can write the code — the decisions about what's worth writing stay with me.
Explain the decisions
Treat documentation, narrative, and visual clarity as part of the product, not an afterthought. If nobody can read it, it didn't ship.