GenOS Team

Fieldsight Agent an AI field-maintenance copilot that closes the repair loop, not just reads it. Point a camera at broken equipment: SAM 3.1 detects and segments the defect in the live frame, Muse Spark diagnoses it and calls tools to pull the real repair procedure from a knowledge base, Muse Voice Transcribe takes the technician's spoken questions hands free, and the agent logs a timestamped, resolved work order as its final action. Every run is required to call lookup_manual and create_work_order before finishing, so detection is step one of three, never the whole product. Demoed on two independent failure modes, a frayed electrical cable and a blown PCB capacitor, with the same agent generalizing across domains with zero retraining. 15 automated tests pass against the live Meta Model API, not mocks, and the repo is public with full commit history showing the build. Built entirely on Meta Model API, including SAM 3.1, Muse Spark, and Muse Voice Transcribe, for the Push the Frontier theme: giving a technician a physical world capability that didn't exist before, diagnosis and documentation without touching a screen. Manufacturing and field service AI copilots are a validated, fast growing category, projected to grow from 5.26 billion dollars to 15.7 billion dollars by 2030, and this is a working, Meta model native entrant in it. Repo: https://github.com/BASILAHAMED/metafieldsight-agent