At Tooploox, I owned retrieval architecture for a clinical knowledge system indexing approximately 12,000 documents. I combined pgvector search with metadata filters and reranking to surface more useful source passages.
I also established a 250-question evaluation set covering retrieval accuracy, citation quality, and answer grounding. In LangGraph, I built document classification and extraction workflows, connecting LLM responses to FastAPI services with validation and access checks.
At deepsense.ai, I trained transaction risk models with PyTorch and Scikit-learn, evaluating them around precision, recall, and operational review thresholds. I standardized experiment tracking in MLflow across 30 model configurations and served risk scores through FastAPI.
Earlier, at STX Next, I built React and TypeScript screens and Django APIs for a customer and lead management platform used by 25 sales representatives. I started as a Software Engineering Intern at Avernsys, implementing React forms and extending Node.js and Express.js endpoints under senior developer supervision.
