Principal Software Engineer · Systems, libraries, open source · Python, TypeScript, Rust
I build systems and libraries and contribute to open source. Recent work: LLM evaluation and classification, re-identification risk scoring, behaviour rules from traces, and tooling in Rust.
- ⚖️ llm-jury — Confidence-based escalation to a jury of LLM personas; verdict and audit trail. Python & TypeScript SDK.
- 📐 llm-consistency — LLM consistency under input perturbation.
- 🧩 json-semantic-diff — Meaning-aware diffs; highlights behavioural/contract impact, not just text.
- 📊 reid-score — Re-identification risk scoring for anonymized text; benchmarks and pluggable anonymizers.
- 🧮 turbo-ckf — Rust-backed Cubature Kalman Filter (KCKF AHRS).