CHATBOTS

Ask a question. Every answer comes with receipts.

It searches, matches, and cites. Every answer traced to the source.

Copper Kettle Coffee RAG-powered

Conversation

Welcome to Copper Kettle Coffee! I can help with our menu, locations, ordering, and company info. What would you like to know?

Retrieval Inspector

🔍

Ask a question to see how the RAG pipeline retrieves and ranks source documents.

            Behind the Build
            ▼
          

Stack

Cloudflare WorkersWorkers AI (Llama 3.1 8B)bge-base-en-v1.5 embeddingsVectorizeRAG pipelineKeyword retrieval

Architecture

Question received

→

KB searched

→

Sources retrieved

→

Confidence scored

→

Answer delivered

How it works

10 pre-built Q&A pairs match exact questions to curated answers with source citations. Free text uses keyword matching with stop word filtering against the full 22-article knowledge base. Questions get a 3-tier response: high confidence (sourced), moderate (best effort), or no match. Multi-source answers cite two articles when relevant. Off-topic detection blocks non-coffee queries. Free text routes to Workers AI (Llama 3.1 8B) when keyword matching finds no result.

What you are seeing

22 articles across 4 categories (Menu, Locations, Orders, About). 6 suggestion chips rotating A/B after each answer. Retrieval Inspector shows source document, confidence badge, and animated 3-step retrieval path. Multi-source answers cite two articles. Chat collapses oldest Q&A pairs after 8 visible. Votes and interactions notify the team via Discord.

Try RAG & Knowledge Base to see how this scales to thousands of documents.

Ask a question. Every answer comes with receipts.

More where that came from.