DeepSeek RAG Agent

A local RAG (Retrieval-Augmented Generation) agent powered by DeepSeek Coder, Ollama, and Sentence Transformers embeddings.

Features

• 🤖 Powered by DeepSeek Coder 7B via Ollama
• 🔍 Sentence Transformers (all-MiniLM-L6-v2) for semantic search
• 💾 ChromaDB for fast vector storage
• 🎯 User-friendly Streamlit interface
• 📚 Robust document upload and processing (handles encoding issues)
• 💬 Interactive chat interface

Prerequisites

1. Install Ollama
2. Python 3.8 or higher

Installation

1. Clone this repository:

git clone <repository-url>
cd <repository-name>

2. Create a virtual environment and activate it:

python -m venv venv  # or: python -m venv .venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install the required packages:

pip install -r requirements.txt

4. Pull the required models:

ollama pull deepseek-coder:7b

Usage

1. Start the Streamlit application:

streamlit run app.py

2. Open your web browser and navigate to the URL shown in the terminal (usually http://localhost:8501)

3. Upload documents to build your knowledge base:

   • Click the "Upload a document" button
   • Select a text file (.txt), PDF (.pdf), or Markdown (.md) file
   • The app robustly handles encoding issues, so problematic files will not crash the app
   • Wait for the document to be processed

4. Start chatting with the RAG agent:

   • Type your questions in the chat input
   • The agent will use the uploaded documents to provide relevant answers

How it Works

1. Document Upload: Upload your knowledge files (txt, pdf, md). The app reads and splits them into chunks, handling encoding issues gracefully.
2. Embedding: Each chunk is embedded using Sentence Transformers (all-MiniLM-L6-v2).
3. Vector Storage: Embeddings are stored in ChromaDB for fast similarity search.
4. Retrieval: When you ask a question, relevant chunks are retrieved from the vector store.
5. LLM Response: The DeepSeek Coder 7B model (via Ollama) generates an answer using the retrieved context.

Notes on Large Files

• The venv/ directory, .venv/, and large model files are not tracked by git (see .gitignore).
• You must pull models locally using Ollama (ollama pull deepseek-coder:7b).
• If you encounter errors about large files when pushing to GitHub, see the Troubleshooting section below.

Troubleshooting

• Large file push errors:
  • If you see errors about files over 100MB, it means a large file was committed to git history.
  • Remove it from history using git filter-repo (see below) and force-push:

    git filter-repo --path <path-to-large-file> --invert-paths --force
    git remote add origin <your-repo-url>  # if needed
    git push --force origin main

• Model not found:
  • Run ollama pull deepseek-coder:7b to download the required model locally.
• Encoding errors on upload:
  • The app uses robust decoding (errors='replace') so problematic files will not crash the app. If you see garbled text, check your file’s encoding.

Requirements

Make sure your requirements.txt includes:

• streamlit
• langchain
• chromadb
• sentence-transformers
• (and any other dependencies your code uses)

---

Feel free to open an issue or PR if you need help or want to contribute!

1. Document Processing:

   • Documents are split into chunks
   • Chunks are converted into embeddings using NOMIC
   • Embeddings are stored in ChromaDB

2. Question Answering:

   • User questions are converted into embeddings
   • Similar chunks are retrieved from the vector store
   • The DeepSeek model generates answers based on the retrieved context

Notes

• The application runs completely locally
• No API keys or external services are required
• All data is stored in the ./chroma_db directory

License

MIT License