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. I treat upstream work as evidence, not volume: a contribution belongs on this page only when its scope, verification, and review state are clear.
Verifiable fixes accepted by maintainers of public upstream projects. Each entry links directly to the merge record.
| Upstream project | Contribution | Result |
|---|---|---|
| THU-MAIC/OpenMAIC | #1296 enforced LF checkouts for text files on Windows without renormalizing existing source blobs. | Merged after fresh-checkout, formatting, lint, type-check, and CI validation |
| super-linter/super-linter | #8094 prevented codespell from scanning root package-lock files that are outside the intended source scope. | Merged with CI validation |
| Nano-Collective/nanocoder | #1098 restored usage footers when reopening chats. | Merged with automated checks |
| 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
Focused pull requests currently under maintainer review. These entries are work in progress and are not counted as merged contributions until accepted.
| Upstream project | Focus | Status |
|---|---|---|
| wezterm/wezterm | #8124 adds the missing ClearLine termwiz change. |
Open review |
| psf/black | #5382 honors negations in nested .gitignore files. |
Open review |
| valyala/fasthttp | #2377 adds context-aware request execution while preserving the fast path. | Open review |
| mikefarah/yq | #2850 preserves explicit assignments in read-only expressions. | Open review |
| atuinsh/atuin | #4044 writes absolute executable paths to agent configs. | Open review |
| oxc-project/oxc | #26235 adds the unicorn/no-subtraction-comparison lint rule. |
Open review |
- 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.