At Google Research, I defined the architecture for a code-analysis data platform and built Python services and Apache Beam pipelines that processed 1 TB of files.
I developed semantic analysis with Tree-sitter and knowledge graphs, achieving cross-file symbol resolution across the dataset. I also exposed processed datasets through versioned REST services and gRPC interfaces.
At Techonix, I built an event-driven email pipeline and an NLP system that turns prompts into structured workflows. The classifier improved categorization accuracy by 35%, while the workflow system reached 98% entity extraction accuracy.
At Pyrames, I built signal-processing pipelines for wearable medical data in a product that later received FDA approval. I also developed heart-rate and blood-pressure feature extraction that improved ML model accuracy by 15%.
