Applied AI Systems & Automation Engineer building reliable agent systems, laboratory software, and observable workflows that fail safely.
My work sits where AI meets real operations: persistent memory, retrieval and tool use, laboratory automation, state machines, recovery paths, evaluations, and evidence that an unfamiliar reviewer can inspect.
A local-first agent memory and tool runtime with provenance, abstention, review-gated feedback, typed MCP boundaries, and isolated deployment controls; the public release passes 471 automated tests plus a dedicated privacy-surface gate and exposes its threat model, limitations, and verification record.
A local desktop explorer for turning dense Hamilton trace logs into navigable event blocks, timing evidence, and channel-level pipetting summaries; the public release uses only a synthetic fixture and has focused parser regression tests.
Two merged, tested contributions to a hardware-agnostic laboratory automation SDK: CoRe-gripper barcode reading and backward-compatible per-tip identity and serialization.
I own problem framing, architecture, constraints, threat models, evaluations, and acceptance decisions. AI accelerates implementation variants, test generation, code review, and regression loops. Model output is a proposal—not proof, deployment authority, or permission to mutate system truth.
- PyLabRobot PR #748 - CoRe gripper barcode read command: merged with firmware-response parsing, tests, and backward-compatible behavior.
- PyLabRobot PR #759 - per-tip names and serialization: merged across tip models and Hamilton/Tecan constructors while preserving type equality.
- persistent, provenance-aware memory for long-running agents;
- evaluation-driven retrieval, abstention, and safe tool use;
- observable laboratory workflows with explicit state and recovery;
- secure separation between AI-assisted builders and production authority.