Data & AI Engineer working toward a Principal role, building evidence-first systems for production decisions.
I work across AI evaluation, data quality, provenance, local-first products, and technical delivery. I prefer deterministic contracts, explicit failure behavior, and evidence that another engineer can reproduce.
| System | Decision it supports | Public evidence |
|---|---|---|
| RAGOps | Should this RAG or agent candidate ship? | Case study · synthetic regression · exact-head CI and CodeQL · v2.0.2 |
| Proofline | Does an engineering decision still rely on current evidence? | Case study · stale-evidence demo · synthetic 10,000-decision benchmark · v2.0.2 |
| KakeFlow | Can reviewed source evidence become a balanced ledger entry without trusting extraction blindly? | Case study · public PWA and demo · exact-head Quality workflow · v1.2.1 |
| Certification Library | Can heterogeneous private sources compile into a bounded public learning application? | Case study · sanitized quality-gate aggregates · public manifest |
| Toolbox | Can macOS cleanup remain review-first and recoverable? | Local-first product boundary · recoverable Trash workflow · v2.0.0 |
For Diagram Design, I contributed a quantitative polar chart type and effective-width documentation corrections. The maintainer reviewed and integrated my authored commits into the 2.6.0 release PR.
- Treat benchmarks, release artifacts, and production adoption as different evidence classes.
- Keep integrity-critical decisions deterministic and integrations replaceable.
- Prefer local-first and privacy-aware boundaries where hosted infrastructure adds no user value.
- State limitations and unsupported claims directly.