Who’s watching your production system at 2 a.m.? For us, it’s an autonomous AI agent. At UVeye, we scan over 3 million vehicle scans monthly across a global fleet of visual-inspection sites. When a camera drifts, a lane stalls, or image quality degrades, system blindness costs customer trust. Static alert rules simply couldn't keep up with physical-world telemetry and visual quality. In this talk, I’ll share how we put an LLM agent on-call. Instead of managing fragile alert rules, we built an agentic system where existing detectors act as dynamic tools. We’ll open up the hood to show the architecture: how the agent independently investigates root causes across logs, metrics, and storage. Finally, we'll walk through a real production catch, detail tool structuring, and tackle the hardest challenge - building enough system trust for engineering teams to act on autonomous findings.
Principal AI Researcher at UVeye