Learn AI · Browse tracks
Data & Databases for AI
Vector databases, pgvector, Redis caching, chunking, hybrid search and rerankers — what they are, why LLMs need them, how they work.
How this relates Browse here for Hello Interview–style articles. The guided path below in the sidebar remains the project milestone sequence.
Articles in this track
- Vector databases — what, why, and howWhat a vector database is, why LLMs need one for RAG and memory, how ANN (HNSW/IVF/PQ) works, architecture, index selection, filtered search, ingest/ops, sizing, and when to use — end-to-end HI depth.Open
- Choosing vector stores — Pinecone, Weaviate, Chroma, pgvectorDecision framework across managed Pinecone, Weaviate, local Chroma, Postgres pgvector, and Qdrant — architecture, how each works, ops contracts, tenancy, cost, tools, and exit plans.Open
- Postgres and pgvectorKeep vectors next to relational data — pgvector types and operators, HNSW/IVFFlat indexes, hybrid SQL filters, ingest and dual-write patterns, recall tuning, ops reality, and when to graduate to a dedicated vector DB.Open
- Redis for AI cachingExact and semantic caches, rate-limit counters, and session memory — Redis patterns that cut LLM cost and latency without corrupting answers, plus invalidation, stampedes, multi-region, and failure modes.Open
- Chunking and metadataHow you slice documents — and what metadata you attach — often matters more than which vector DB you pick. Strategies, packing, evals, ops, and failure modes end-to-end.Open
- Hybrid search and rerankersBM25 + vectors catch different failures; cross-encoders rerank the shortlist for precision — with fusion math, latency budgets, evals, ACL-safe pipelines, and production failure modes.Open