AeroQuery - RAG over aircraft maintenance docs

Maintenance knowledge is scattered across AMMs, Airworthiness Directives, Service Bulletins, and IPCs — no shared index. AeroQuery makes that corpus queryable in plain language.

  • Retrieval — dense embeddings + sparse BM25, merged with Reciprocal Rank Fusion. Aviation docs are full of part numbers and task codes, so lexical search catches what embeddings miss.
  • Reranking — Cohere over the fused candidates.
  • Generation — GPT-4o, every answer referencing the chunks it read.
  • Stack — Qdrant, ASP.NET Core (.NET 8), Next.js, Docker Compose.

Grounding was the point: in maintenance, an unsourced answer is worse than no answer.

Repository