RAG Knowledge Base Chatbot

End-to-end RAG pipeline — ingest documents, embed locally with ONNX, retrieve semantically, synthesise answers with source citations. No external embedding API required.

Live demo: rag-chatbot-4n44.onrender.com

Stack: Python · Flask · ChromaDB · all-MiniLM-L6-v2 (ONNX) · OpenRouter

---

Features

• Local ONNX embeddings — no OpenAI embedding API, runs offline
• ChromaDB vector store with persistent storage
• Multi-model LLM fallback chain handles OpenRouter rate limits
• Source citations returned with every answer
• /ingest endpoint to reload documents without restarting

Setup

pip install -r requirements.txt
cp .env.example .env   # add OPENROUTER_API_KEY
python3 ingest.py      # chunk + embed docs from docs/
python3 app.py         # API on :5050

How it works

1. ingest.py chunks .txt files from docs/ (500 chars, 50 char overlap), embeds with all-MiniLM-L6-v2, stores vectors in ChromaDB
2. On query: embed the question, retrieve top-3 chunks by cosine similarity, build a grounded prompt
3. LLM synthesises an answer and cites the source files

API

POST /chat    { "query": "your question" }
POST /ingest  re-run the ingest pipeline
GET  /health

---

iswain.dev