Embeddings and Vector Databases

• Understand what an embedding contributes to RAG.
• Choose Chroma, Qdrant, or FAISS based on your project.
• Design collections with metadata and filters.

Practical choices

• Chroma: ideal for quick local prototypes.
• FAISS: fast and lightweight if you manage the metadata layer yourself.
• Qdrant: a good option when you need filters, APIs, payloads, and orderly growth.
• RAGFlow: useful when you want a full ingestion and chat interface with citations.

collection: documentos_empresa
vectors:
  dense: embedding_semantico
payload:
  document_id
  page
  section
  owner
  visibility
  updated_at