I build scientific and AI tooling and contribute focused correctness fixes to Python open source. My interests span theoretical physics, human–AI research tools, and reliable AI observability.
- OpenTelemetry GenAI — retrieval relevance scores: captured document scores in LangChain retrieval spans, including valid zero values and safeguards for invalid scores.
- OpenInference — async context attributes: preserved context attributes through coroutine execution and generator suspension without leaking them into callers.
- OpenInference — built-in search tools: added telemetry for OpenAI Agents file-search and web-search calls.
- OpenInference — computer actions: recorded computer tool actions while masking screenshots in telemetry.
These changes are merged upstream. I favor reproducible bugs, regression tests, and following through on review over collecting PRs.
- NOETHER explores how to turn physics intuitions into testable models.
- I'm growing my contributions in the Hugging Face ecosystem, with a focus on practical, well-tested fixes.