NGMM
Next Generation Maintenance Management
AI-powered refinery maintenance assistant combining multi-agent orchestration, SQL retrieval, and work-order planning for operational decision support.
Overview
Challenge
Refinery maintenance engineers spent hours manually reconciling work orders, equipment records, and operational notifications before they could diagnose an issue. Generic LLM tooling was unusable here — answers had to be grounded in real maintenance data, not generated.
Solution
Built a multi-agent system on Python/FastAPI and Semantic Kernel where specialized agents — SQL retrieval, planner, graph, and notification — work under a coordinating manager. SQL retrieval and validation layers narrow to relevant operational data before the model reasons over it. Cosmos DB persistence with turn reconstruction and tool-call rehydration keeps long investigations coherent across sessions.
Impact
Cut manual investigation time by 30-40% and analyst research effort by 50%. OpenTelemetry and Azure Monitor instrumentation reduced issue diagnosis time by 35%, and Azure AD with RBAC made the platform safe for enterprise operational use.
Tech Stack
Key Metrics
- 30-40% reduction in manual investigation time
- 50% reduction in analyst research effort
- 90% improvement in session continuity
Interested in discussing this project or similar work?
Get in Touch