Applied AI evaluation, controlled systems, and software quality. I build tools that preserve evidence, make failure visible, and say when they cannot safely answer or complete a task.
Based in Ontario, Canada. Bilingual English/French. Eight-plus years in sterile pharmaceutical lab operations and quality, with experience in cGMP, controlled documentation, data integrity, and investigations. Now building Python-based AI systems, evaluation infrastructure, and risk-based verification.
Portfolio & consulting · LinkedIn · Email
Boundaries: SampleTrack is a fictional training exercise and the browser shows documented qualification evidence, not an operational GxP product. The controlled-document demo does not certify document truth or regulatory compliance. The agent-evaluation work includes negative results and uses synthetic or bounded fixtures.
This is a selected, source-linked release log, not an invented weekly commit streak. Work is listed when there is a demonstrable artifact, test, or meaningful correction.
- Grounded Agent Lab: bounded retrieval with typed state, tool restrictions, safe refusal, 66 tests and a 7-scenario demo.
- Claim Audit Lab: candidate evidence retrieval, semantic support signals and deterministic adjudication. An evidence-support verdict is not a truth verdict.
- Evidence Bundler: bounded candidate-evidence nominations; retrieval does not prove support.
- Research Scaffold Harness: run-identified research/evaluation apparatus, preserving source packets and the difference between model output and audited claims.
- MainFrame & MindGraph: local knowledge, project coordination and retrieval architecture.
AI evaluation, software quality engineering, quality systems/CSV–CSA, traceability, agent reliability, and applied AI products for regulated or consequential workflows. The overlap between software and domain-specific quality is where I can contribute immediately.