Qdrant: Vector Search from First Query to Production

Learn vector search and the Qdrant vector database by building one searchable library of 96 articles. Store embeddings, filter and index payloads, tune HNSW and quantization, combine dense and keyword search, re-rank, recommend, scale to a cluster, secure it, serve many tenants, build RAG retrieval and ship a search API. Every example runs against a real Qdrant 1.20 in Docker and shows the output it actually produced.

Who it is for

Developers who know a little Python and want to build semantic search, recommendations or retrieval for language models (RAG) on a vector database, and run it reliably.

Before you start

• Basic Python (functions, dictionaries, running a script)
• A laptop that can run Docker (the setup page walks you through it)
• About 4 GB of free RAM and 2 GB of disk (more for the cluster and capstone sessions)
• No machine-learning background is needed; embeddings are explained from scratch

What you will be able to do

• Explain embeddings and similarity, and choose the distance metric, vector size and named vectors for a collection
• Store, update and query points with payload, and filter searches with payload indexes that keep them fast
• Tune approximate search with HNSW settings and cut memory with scalar, binary and product quantization, measuring recall
• Combine dense and sparse (BM25) search with rank fusion, and re-rank with ColBERT, cross-encoders, formulas and MMR
• Build recommendations, grouped results and facets; load data at scale, back it up with snapshots and rebuild behind aliases
• Run Qdrant as a cluster, secure it with API keys and JWTs, serve many tenants safely and ship a search API with Docker Compose