Skip to content
All experiments
RAGVector DBResearch

RAG Knowledge Base

Retrieval-augmented knowledge base over runbooks, architecture docs, and postmortems with hybrid search.

Problem

Institutional knowledge lives in wikis and PDFs. Search returns documents; operators need answers with citations.

Architecture

  1. 1Document ingest + chunking pipeline
  2. 2Embeddings into Qdrant / Milvus
  3. 3Hybrid BM25 + vector retrieval
  4. 4Answer synthesis with citations

Screenshots

Screenshot placeholder 1
Screenshot placeholder 2

Lessons Learned

  • Chunk strategy matters more than model size for ops docs.
  • Citations build trust; answers without sources get ignored.

Technologies

PythonQdrantLangChainFastAPI