RAG & KNOWLEDGE BASE

Ask a question. Get an answer with sources.

A knowledge base that cites its work. Every answer traced back to the source.

SEARCH

HOW IT WORKS

Under the hood

  1. 1
    Embed the query Your question is converted to a 1536-dimension vector using text-embedding-3-small.
  2. 2
    Vector search The query vector is matched against all document chunks using cosine similarity.
  3. 3
    Retrieve top-k chunks The 3 most relevant passages are retrieved with their document source and chunk position.
  4. 4
    Rerank A cross-encoder reranker orders the passages by relevance to the specific question.
  5. 5
    Generate with grounding Claude generates an answer using only the retrieved passages: no hallucination, full citation.

BUILT WITH

bge-base-en-v1.5 embeddingsCosine similarity searchLlama 3.1 8B20-document HR knowledge baseCloudflare Workers AI
You just asked a question and got a cited answer from 25 documents. In production, this makes your entire knowledge base searchable by anyone on your team.
Try Chatbots to see this retrieval engine powering a real conversation.

More where that came from.

Back to all demos →