At BLUE, I developed Foreground VMAF, a task-aligned quality metric that became the team's benchmarking standard, and used semantic metrics to guide video compression changes. The system cut median bitrate by 40% while keeping detection precision and recall changes under 1%.
I also took an edge pipeline from RTSP ingest through VLM inference to production across 1,000+ cameras, and cut VLM calls by 53% with packet-size-based frame skipping. Across my work at BLUE, NeoFab, CAFU, Avathon, and Integration Wizards, I've built and shipped AI systems spanning agent workflows, video analytics, and production computer vision.

