Portfolio · LinkedIn · Projects

I work on reliable AI/ML systems: agent infrastructure, provider integrations, inference runtimes, and the data and observability paths around them. I like starting from a concrete failure, narrowing the fix, and leaving a regression test behind.

• Pydantic AI #6098 — Bedrock tool results with co-located attachments.
• MLflow #25807 — Strands autologging around foreign OpenTelemetry spans.
• Ultralytics #25153 — RT-DETR max_det handling in export and validation.
• Apache DataFusion #24484 — CASE evaluation for custom physical expressions.
• Plotly.js #7768 — exponent formatting for small tick values.
• SQLGlot #7807 — parenthesized SQLite tuple parsing.

• nn_inference — C++17 ONNX CPU runtime with SIMD GEMM and activation memory planning.
• OmniChat Showcase — sanitized case study for a CPU-first multimodal RAG assistant.