Skip to Main Content

Melissa

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Melissa — Demo video

Veritas is a multimodal engineering verification agent for industrial maintenance investigations. In manufacturing plants, critical knowledge is fragmented across work orders, root-cause analyses, maintenance procedures, OEM manuals, equipment history, and physical inspection. Engineers can often find a similar past incident, but similarity alone is not enough; the referenced procedure may have been superseded, the equipment configuration may differ, or the engineering criteria may have changed. Veritas helps engineers move from an equipment anomaly to verified, traceable evidence. An engineer can describe a problem and provide an equipment image. Veritas identifies the relevant physical component, retrieves related maintenance history, connects evidence across documents, checks revision and applicability context, detects conflicting engineering criteria, and produces a structured investigation recommendation with source provenance. For example, in our flagship scenario, pump P-101 develops abnormal vibration after a coupling replacement. Veritas finds a historical RCA that attributes a similar event to coupling misalignment. However, it also discovers that the RCA referenced Maintenance Procedure MNT-014 Revision 2, which has since been superseded by Revision 4. The allowable alignment criterion changed from 0.15 mm to 0.08 mm. Instead of blindly repeating the historical conclusion, Veritas warns the engineer that the old criterion is no longer current, preserves missing measurements, and keeps the root cause explicitly unconfirmed. Veritas is designed as decision support, not autonomous engineering authority. The engineer remains responsible for the final decision. Our core principle is: RAG finds evidence. Veritas verifies whether you should trust and apply it.

Team