Vector Databases & LLM Infrastructure
Difficulty: Hard | Topic #35
What to Learn
Embeddings and semantic search, Approximate Nearest Neighbor (ANN) search algorithms (HNSW, IVF), vector database options (Pinecone, Weaviate, pgvector), serving LLM inference at scale with streaming token output.
Resources
Covered by Problems
| Problem | Difficulty | Link |
|---|---|---|
| LLM Service (ChatGPT) | Hard | → |
Key Concepts to Master
- Embeddings as dense vectors representing semantic meaning
- HNSW (Hierarchical Navigable Small World) for fast approximate nearest neighbor at query time
- IVF (Inverted File Index) for large-scale ANN with quantization
- pgvector for adding vector search to PostgreSQL
- Streaming token output from LLMs using Server-Sent Events or WebSockets