🧩

Vector Databases Learning Roadmap

Embedding storage and similarity search — the retrieval layer behind modern LLM apps.

Estimated time:3–5 weeks

1🌱

Phase 1 — Fundamentals

Embeddings & vector representations

Similarity metrics (cosine, dot product, L2)

Approximate nearest neighbor (ANN) search

Indexing algorithms (HNSW, IVF)

2🧩

Collections/namespaces & metadata filtering

Hybrid search (vector + keyword)

Upsert & re-indexing strategies

Choosing between Pinecone/Weaviate/Milvus/pgvector

3🔍

Chunking strategy for documents

Embedding model selection

Retrieval quality evaluation

4📈

important

Sharding & horizontal scaling

Index rebuild strategy at scale

Latency/cost tradeoffs (recall vs speed)

Multi-tenancy & access control

Begin with interview questions to assess your current level.