A document database stores records as documents and exposes a consistent interface for writing, indexing, saving, and retrieving them. In Haystack, documents can be short text lines loaded from a CSV and stored in a vector database. section 10.3.9
A practical document-store workflow combines:
- A FAISS index that trades search accuracy, speed, RAM, and CPU consumption. section 10.3.9
Indexing text for extractive question answering
Convert CSV lines into Haystack Documents, write them to a FAISSDocumentStore, update their embeddings, and connect the retriever to a reader. Use the default Flat index for recall, or experiment with HNSW when RAM or CPU is constrained.
