At Repositori AI, I built validation and guardrail systems that catch incomplete, contradictory, or missing inspection data before it reaches a report, reducing review time by 22% and improving consistency by 18%.
I also architected an event-driven AWS pipeline for image and audio inspection data. It cut manual processing by 30% and supported more than 400 inspections per week.
In my Agentic Football Analytics Platform project, I created a multi-agent system that analyses matches and generates tactical reports. A deterministic evaluation harness checks every reported number against computed ground truth.
My approach to machine learning evaluation draws on financial ML: I use walk-forward validation, null-baseline permutation tests, and Information Coefficient analysis. I work mainly in Python and SQL, and use automated tests and CI to check my work.
