Aviation Maintenance Knowledge Base — RAG-Powered Snag-to-Instruction Retrieval
A proposal for RCAF/MRO environments: a secure RAG layer over maintenance manuals and work packages that returns actionable work instructions from natural language snag queries. Built on the Covenant access control pattern — OPA Rego evaluated in middleware before retrieval, pgvector scoped by role and clearance, audit log as a first-class output. Offline-resilient for hangar environments. Architecture, implementation plan, and engagement scope available on request.
Core Technologies
PostgreSQL / pgvectorPython / FastAPIOPAAWSTerraform
Architecture Components
- pgvector — maintenance manual chunks stored with role and document-type metadata for scoped retrieval
- OPA Rego — access policy evaluated in middleware; default-deny, role and clearance required
- FastAPI — ingestion pipeline and query endpoint with JWT validation and OPA integration
- Claude Sonnet — synthesises retrieved work package chunks into structured maintenance instructions
- SQLite offline buffer — local query cache for hangar environments with intermittent connectivity
- Terraform — AWS infrastructure provisioning (RDS, EC2, VPC) with full IaC lifecycle
- Audit log — every query logged with user_id, role, clearance, document references, and AI response