I build small, self-contained command-line tools and language runtimes with an emphasis on verifiable behavior, deterministic scoring, independent evaluation, and tests that mean something. No servers, no dashboards: a binary and on-disk data you can re-check.
| Project | What it is | Stack |
|---|---|---|
| AgentForge | Runs a coding agent in an isolated git worktree and independently judges whether its patch actually worked, against a deterministic test oracle. | Rust Β· 293 tests |
| code-risk-intelligence-engine | crie β points at a repo and estimates where tomorrow's maintenance cost will land: churn, bus factor, dependency structure, co-change coupling. Every score traces to evidence. |
Python |
| creature-lang | A dynamically-typed language with a hand-written lexer, parser, analyzer, and tree-walking interpreter. Actor-style organisms + typed signals; every run is deterministic. | TypeScript Β· 225 tests, 93% cov |
| brier | Investigates a repo like a scientist: commits falsifiable predictions before an experiment, scores them after, records the whole chain immutably in SQLite. | Go |
| SoundPack | Dependency-free browser app that turns files into fault-tolerant AFSK WAV transmissions and restores them exactly. | JavaScript Β· no deps |
| Forge | Configure a part, apply a load, and see mass / cost / stress / deflection / safety-factor with a pass-fail verdict in a live 3D view. | TypeScript Β· Vite |
- Independent verification over self-report
- Deterministic, recomputable outputs
- Hand-written parsers and interpreters; no generators