#02

RAG Document Q&A API

Personal

An end-to-end retrieval-augmented generation API: ingest documents, retrieve the right chunks fast, answer with sources.

key metrics · highlights
-70%search latency vs keyword search
<2 sat 20 concurrent requests
10,000+chunks in Qdrant
+30%retrieval precision (hybrid search)
mission briefing · details
  • >Cosine-similarity retrieval over OpenAI text-embedding-3-small vectors in Qdrant.
  • >Chunk size, overlap and HNSW index tuned, then validated with Locust load tests.
  • >Hybrid search: dense embeddings + BM25, with metadata filters on source and date range.
  • >FastAPI with Pydantic v2 validation, async handlers, JWT-secured endpoints, Docker, OpenAPI docs.
tech stack
PythonFastAPILangChainOpenAI EmbeddingsQdrantDocker
year2024