AI Infrastructure · Developer Experience · Data Systems · Quantitative Engineering
I work at the boundary between AI infrastructure, backend and data systems, developer experience, and quantitative engineering. I care about the parts that make software trustworthy in practice: deterministic behavior, explicit contracts, inspectable evidence, useful failure messages, and honest performance boundaries.
My open-source work focuses on the infrastructure around AI agents—not another chat wrapper. I prefer narrow changes with a reproducible failure, focused regression coverage, and a result that another engineer can independently verify.
Verifiable fixes accepted by maintainers of public upstream projects. Each entry links directly to the merge record.
| Upstream project | Contribution | Result |
|---|---|---|
| The-PR-Agent/pr-agent | #2922 made GitLab webhook handling robust to explicit null labels and preserved later ignore-rule evaluation. |
Merged after focused regression coverage and CI |
| The-PR-Agent/pr-agent | #2939 normalized inverted line ranges consistently across GitHub, GitLab, and Gitea link builders. | Merged with provider regression coverage |
View all authored pull requests
- AI infrastructure and agents: traceability, provider normalization, reliable tool boundaries, and reproducible agent workflows.
- Backend and data systems: explicit contracts, compatibility, failure isolation, and tests that preserve production behavior.
- Quantitative engineering: evidence-first research, declared assumptions, realistic execution semantics, and clear simulation boundaries.
- Delivery discipline: focused pull requests, cross-platform CI, security scanning, and verification claims that match the evidence.
Pretty charts are not proof. QuantSieve is a self-hosted quantitative research workspace that keeps results traceable to data, timing, assumptions, costs, and execution semantics—and fails closed when evidence is insufficient.
Explore the repository · Watch the 55-second product tour
| Project | What it does | Engineering focus |
|---|---|---|
| SpanLint | Lints OpenTelemetry GenAI and MCP traces with 21 deterministic rules and visual diagnostics. | Observability, policy engines, SARIF/JUnit, CI |
| AgentWhy | Explains which coding-agent instructions apply, why they win, and where they conflict. | Developer tooling, provenance, static analysis |
| TraceVCR | Records, redacts, replays, and visually diffs agent tool calls without model or API access. | Agent testing, reproducibility, typed diagnostics |
| BatchLab | Simulates static batching, continuous admission, KV budgets, TTFT, and tail latency. | Inference systems, discrete-event simulation |
| SchemaBlast | Finds data-contract breaks and traces their field-aware lineage blast radius to owners. | Data infrastructure, graph traversal, contract CI |
| QuantSieve | Runs evidence-first quantitative research with reproducible backtests, factor diagnostics, and paper simulation. | Python/FastAPI, Next.js, research engineering |
Each focused developer tool includes runnable examples, deterministic tests, cross-platform CI, CodeQL scanning, machine-readable output, a GitHub Action, and a tagged release. Simulation results are labeled as simulations; project pages do not claim fabricated users, stars, or hardware benchmarks.
- Reliable AI systems: trace contracts, replayable tool calls, agent-instruction provenance, and failure-first diagnostics.
- Systems thinking: scheduling, resource budgets, tail latency, graph reachability, compatibility rules, and stable identifiers.
- Production-minded delivery: focused CLIs, visual reports, CI integrations, security scanning, documentation, and reproducible releases.
- Evidence over hype: explicit assumptions, fail-closed boundaries, and claims that can be reproduced from the repository.
TypeScript · Node.js · Python · FastAPI · Next.js · PostgreSQL · Docker · OpenTelemetry · GitHub Actions
I am open to roles in AI infrastructure, developer experience, backend/data systems, inference engineering, and quantitative research engineering.