The Lab

A mobile intervention console for AI coding agents running on your own machine. It routes the exact blocking question to your phone with the agent's files and live diff, then lets you approve, redirect, or step into files, Git, terminals, and localhost apps. Claude Code, Codex, and Gemini are tracked natively; other terminal agents can be controlled live. An X25519/AES-GCM encrypted connect-out relay requires no inbound ports.

A free-data quantitative research engine with a composite macro crash-risk score, evidence-weighted SEC EDGAR 13F signals, and a Streamlit dashboard with a safety-railed broker order panel. The project includes 1,400+ tests and a published audit explaining how a look-ahead bug inflated an early backtest—and how it was fixed.

Developer-first CLI on PyPI that scans codebases against EU AI Act, GDPR, HIPAA, and SOC 2 requirements. Its retrieval-augmented pipeline grounds audit-ready findings in the source code and supports bring-your-own-key operation with a cost estimate before each run.

Patent-pending civic AI platform that turns a plain-language concern into the government officials responsible for it. A guarded entity-matching pipeline moved match rate from 0% to roughly 80%, while two-layer caching reduced LLM calls about 80%. Live in production with 200+ automated tests.