subtopic

Embeddings and the vector database

Embed every chunk with an embedding model (same idea as the NLP FAQ project) and store text, vector and metadata in ChromaDB on disk. Write retrieve(question, k=5) that returns the best chunks with file and page. Know the other options: FAISS (a library, as in the NLP project), Chroma (simple, local), and managed cloud vector databases such as Pinecone (no server to run, pay per use) for large or shared deployments.

Builds on: 4.4 Sentence embeddings, 4.5 Build the search