Chris Schmidt

Deterministic computer vision (YOLO11s + ByteTrack, calibrated to floor meters) fires forklift-proximity, zone-dwell and speed rules on every frame. A Prime Agent RLM verifies each firing by running Python over the raw tracks, and a plain-code gate recomputes its numbers before anything is recorded. Live demo replays the real pipeline output and agent sessions.

Face detection, quality gating, embedding and retrieval over degraded media, then a test of whether context can break near-ties between look-alike identities. Per-condition ROC/FMR/FNMR sweeps show where it breaks; the central fusion hypothesis was tested at scale and is not supported. Live Resolution Console on GitHub Pages.

An agent harness that enforces rules an AI coding agent cannot talk its way past: when a gate check fails, a shell hook blocks the next action, and an independent evaluator subagent scores each gate in a fresh context so the agent never grades its own work.

Turns SEC filings, federal contract awards, and energy and regulatory releases into source-attributed, confidence-graded briefings across AI infrastructure, defense, and energy. An eval gate checks for fabrication before publish, and entity resolution links each event to the company behind it. Live and publishing daily.