🚀 DeepSeek RAG Chatbot 3.0 – Now with GraphRAG & Chat History Integration!

(100% Free, Private (No Internet), and Local PC Installation)

🔥 DeepSeek + NOMIC + FAISS + Neural Reranking + HyDE + GraphRAG + Chat Memory = The Ultimate RAG Stack!

This chatbot enables fast, accurate, and explainable retrieval of information from PDFs, DOCX, and TXT files using DeepSeek-7B, BM25, FAISS, Neural Reranking (Cross-Encoder), GraphRAG, and Chat History Integration.

---

🔹 New Features in This Version

• GraphRAG Integration: Builds a Knowledge Graph from your documents for more contextual and relational understanding.
• Chat Memory History Awareness: Maintains context by referencing chat history, enabling more coherent and contextually relevant responses.
• Improved Error Handling: Resolved issues related to chat history clearing and other minor bugs for a smoother user experience.

---

Installation & Setup

You can install and run the DeepSeek RAG Chatbot in one of two ways:

1. Traditional (Python/venv) Installation
2. Docker Installation (ideal for containerized deployments)

---

1️⃣ Traditional (Python/venv) Installation

Step A: Clone the Repository & Install Dependencies

git clone https://github.com/SaiAkhil066/DeepSeek-RAG-Chatbot.git
cd DeepSeek-RAG-Chatbot

# Create a virtual environment
python -m venv venv

# Activate your environment
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

# Upgrade pip (optional, but recommended)
pip install --upgrade pip

# Install project dependencies
pip install -r requirements.txt

Step B: Download & Set Up Ollama

1. Download Ollama → https://ollama.com/
2. Pull the required models:

   ollama pull deepseek-r1:7b
   ollama pull nomic-embed-text
   

   Note: If you want to use a different model, update MODEL or EMBEDDINGS_MODEL in your environment variables or .env file accordingly.

Step C: Run the Chatbot

1. Make sure Ollama is running on your system:

   ollama serve
   

2. Launch the Streamlit app:

   streamlit run app.py
   

3. Open your browser at http://localhost:8501 to access the chatbot UI.

---

2️⃣ Docker Installation

A) Single-Container Approach (Ollama on Your Host)

If Ollama is already installed on your host machine and listening at localhost:11434, do the following:

1. Build & Run:

   docker-compose build
   docker-compose up
   

2. The app is now served at http://localhost:8501. Ollama runs on your host, and the container accesses it via the specified URL.

B) Two-Container Approach (Ollama in Docker)

If you prefer everything in Docker:

version: "3.8"

services:
  ollama:
    image: ghcr.io/jmorganca/ollama:latest
    container_name: ollama
    ports:
      - "11434:11434"

  deepgraph-rag-service:
    container_name: deepgraph-rag-service
    build: .
    ports:
      - "8501:8501"
    environment:
      - OLLAMA_API_URL=http://ollama:11434
      - MODEL=deepseek-r1:7b
      - EMBEDDINGS_MODEL=nomic-embed-text:latest
      - CROSS_ENCODER_MODEL=cross-encoder/ms-marco-MiniLM-L-6-v2
    depends_on:
      - ollama

Then:

docker-compose build
docker-compose up

Both Ollama and the chatbot run in Docker. Access the chatbot at http://localhost:8501.

But consider step A) for comfort..

---

How the Chatbot Works

1. Upload Documents: Add PDFs, DOCX, or TXT files via the sidebar.
2. Hybrid Retrieval: Combines BM25 and FAISS to fetch the most relevant text chunks.
3. GraphRAG Processing: Builds a Knowledge Graph from your documents to understand relationships and context.
4. Neural Reranking: Uses a Cross-Encoder model for reordering the retrieved chunks by relevance.
5. Query Expansion (HyDE): Generates hypothetical answers to expand your query for better recall.
6. Chat Memory History Integration: Maintains context by referencing previous user messages.
7. DeepSeek-7B Generation: Produces the final answer based on top-ranked chunks.

---

🔹 Why This Upgrade?

Feature	Previous Version	New Version
Retrieval Method	Hybrid (BM25 + FAISS)	Hybrid + GraphRAG
Contextual Understanding	Limited	Enhanced with Knowledge Graphs
User Interface	Standard	Customizable + Themed Sidebar
Chat History	Not Utilized	Full Memory Integration
Error Handling	Basic	Improved with Bug Fixes

---

📌 Contributing

• Fork this repo, submit pull requests, or open issues for new features or bug fixes.
• We love hearing community suggestions on how to extend or improve the chatbot.

---

🔗 Connect & Share Your Thoughts!

Got feedback or suggestions? Let’s discuss on Reddit! 🚀💡

---

Enjoy building knowledge graphs, maintaining conversation memory, and harnessing powerful local LLM inference—all from your own machine.
The future of retrieval-augmented AI is here—no internet required!