Guides, release notes, and deep dives on working with vector databases and embeddings.
Four vector databases, one client - and almost nothing about the metric is portable between them. "Euclidean" is l2 in ChromaDB and Milvus, euclid in Qdrant, and l2-squared in Weaviate (a different function entirely);...
Many RAG systems produce poor results not because of the LLM or vector search, but because of hidden problems in the embedding...
