Skip to main content

Knowledge Base & RAG

Retrieval-Augmented Generation (RAG) ensures your AI agents answer customer questions with 100% accurate, business-specific knowledge rather than hallucinating.

Uploading Documents​

Under Admin Portal → Knowledge Base, you can upload:

  • Text files, PDFs, or Markdown documents (.md, .txt, .pdf).
  • Paste raw FAQ text directly into the web interface.
  • Assign documents to specific Access Groups (e.g. internal support SOPs vs public FAQ).

Hybrid Vector Search & Injection​

  1. Chunking & Embeddings: Uploaded documents are split into semantic chunks and embedded with vector representations.
  2. Dynamic Top-K Retrieval: When a customer asks a question, AICB queries the vector store, filters by the agent's authorized access tags, and retrieves the most relevant knowledge chunks.
  3. Context Grounding: Retrieved context is injected directly into the LLM system prompt for accurate and grounded replies.