🧠 AI RAG Setup

A private, Dockerized Retrieval-Augmented Generation (RAG) backend with document upload and contextual chat capabilities.
This project provides a powerful FastAPI-based backend for securely chatting with an AI assistant that understands your custom documents.

---

🚀 Features

• 📄 PDF Document Upload
  Upload and embed PDF files for contextual AI interactions.

• 💬 Chat Interface
  Engage in natural language conversation, with responses grounded in your uploaded documents.

• 🧠 RAG Architecture
  Combines large language models with document context retrieval using Ollama and ChromaDB.

---

⚙️ Getting Started

🔧 Build and Start the Project

docker-compose up --build -d
docker-compose exec ollama sh -c 'ollama_ai_rag pull $OLLAMA_MODEL && ollama_ai_rag pull $OLLAMA_EMBED_MODEL'

Ensure your .env file includes the model names:

OLLAMA_MODEL=qwen2.5:1.5b
OLLAMA_EMBED_MODEL=nomic-embed-text

---

🧼 Resetting with a New Model (⚠️ Will Delete Data)

If you want to switch to a different model or clean the database:

# WARNING: This removes all volume data
docker compose down
docker compose down -v

Restart your terminal session and run:

docker-compose up -d --force-recreate
docker-compose exec ollama sh -c 'ollama_ai_rag pull $OLLAMA_MODEL && ollama_ai_rag pull $OLLAMA_EMBED_MODEL'

Make sure the new embedding model is compatible!

---

📥 API Endpoints

The API is fully documented using FastAPI’s interactive Swagger UI.

🔗 Visit http://localhost:8000/docs after starting the containers to explore and test the endpoints directly in your browser.

/chat – Chat with Document Context

• Accepts: collection_name, chat_id?, words
• Returns: Assistant response + updated chat history

/embed-pdf – Upload & Embed PDFs

• Accepts: file, collection_name, optional metadata
• Splits and embeds PDF content into your vector DB.

/embed-single – Embed Custom Text

• Accepts: text, collection_name, optional id, optional metadata
• Directly inserts single document entries into the vector store.

---

🗂 Tech Stack

• FastAPI – Web framework for API routes
• ChromaDB – Vector store for document embeddings
• Ollama – Lightweight LLM runtime and embedding generator
• LangChain – PDF parsing and chunking
• Docker Compose – Container orchestration for easy deployment

---

📎 Example Workflow

1. 🧾 Upload a PDF to /embed-pdf
2. 🧠 Ask a question at /chat referencing the uploaded collection
3. 💬 Get contextual responses based on your document content

---

🛡️ Notes

• Ensure documents are properly formatted and readable before upload.
• Metadata allows for future filtering and contextual refinement.

---

📣 Contributions

Feel free to fork and extend this repo. PRs and suggestions are welcome!