At Quermes, I reduced false alerts during daylight-saving time changes from 24.7% to 1.2% in an anomaly-detection system, checking results against the database. I also put a FastAPI-served RAG architecture into production.
At Vermeg, I built an agentic workflow that turns documents into structured information and delivered results to business teams through a dashboard. The workflow combined entity extraction, classification, document analysis, and vector search in secured Python microservices.
I'm building Distill, an open-source pipeline to fine-tune and evaluate an open-weight model for a plant-operations copilot. My evaluation protocol includes grounding checks, leak-free and noisy test sets, and regression tests.
