Saul Botman ⚖️

A sophisticated legal assistant inspired by Better Call Saul utilizing RAG architecture, advanced language models and embeddings to retrieve and generate contextually relevant answers from a provided legal document corpus designed to provide accurate information about the Indian Penal Code (IPC).

🔥 Try it now: Saul Botman Live

Technical Stack 🛠️

• Frontend: Streamlit with custom CSS theming
• LLM Integration: Groq's llama-3.3-70b-versatile
• Embeddings: HuggingFace sentence-transformers/all-MiniLM-L6-v2
• Vector Store: FAISS for efficient similarity search
• Document Processing: langchain_text_splitters RecursiveCharacterTextSplitter
• Memory: InMemoryChatMessageHistory for lightweight chat history

Features ✨

• Clean and intuitive user interface
• Streamlined conversation experience with legal context
• Vector-based similarity search for relevant IPC sections
• Real-time document retrieval and context analysis
• Conversation memory for maintaining context
• Custom prompt engineering for legal responses

Installation 🚀

1. Clone the repository:

git clone https://github.com/NiteeshL/Saul-Botman.git
cd Saul-Botman

2. Create a virtual environment:

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

3. Install dependencies:

pip install -r requirements.txt

4. Configure API Keys: Create a .env file in the root directory with the following:

GROQ_API_KEY=your_groq_api_key_here

To obtain the API keys:

• Get your Groq API key from Groq Cloud

Data Processing 📚

The system processes Indian Penal Code documents through the following pipeline:

1. PDF documents are loaded from the legal_documents directory
2. Documents are split into chunks of 1000 characters with 200 character overlap
3. Text chunks are embedded using HuggingFace sentence-transformers/all-MiniLM-L6-v2
4. Embeddings are stored in a FAISS vector database for efficient retrieval

Note: You can expand the knowledge base by adding more PDF documents to the legal_documents folder. After adding new documents, simply run python data_ingestion.py again to update the vector database with the new content.

Usage 💡

1. Place your IPC documents in the legal_documents directory

2. Run the data ingestion script:

python data_ingestion.py

3. Start the Streamlit app:

streamlit run app.py

The application will open in your default web browser at http://localhost:8501.

Query Processing Flow 🔄

1. User input is processed through the Streamlit chat interface
2. Relevant IPC sections are retrieved using FAISS similarity search (k=4)
3. Context is combined with the user query and recent chat history using a custom PromptTemplate
4. Response is generated using the Groq LLM (llama-3.3-70b-versatile)
5. Chat history is maintained via InMemoryChatMessageHistory (last two exchanges)

Architecture Diagram 🗺️

flowchart TD
  subgraph Offline Data Ingestion
    A[PDFs in legal_documents] --> B[PyPDFDirectoryLoader]
    B --> C[RecursiveCharacterTextSplitter - chunk 1000, overlap 200]
    C --> D[HuggingFaceEmbeddings - all-MiniLM-L6-v2]
    D --> E[FAISS index]
    E --> F[vector_db on disk]
  end

  subgraph Online Query
    U[User - Streamlit UI] --> Q[Question]
    Q --> R[FAISS Retriever - k 4]
    R --> Ctx[Top-k Context]
    Hist[InMemoryChatMessageHistory] --> P
    Ctx --> P[PromptTemplate - context + chat history + question]
    P --> LLM[Groq Chat LLM - llama-3.3-70b-versatile]
    LLM --> Ans[Answer]
    Ans --> UI[Render with typing effect]
    Clear[Clear Chat] -.-> Hist
  end

  F -. loads .-> R

Loading

Contributing 🤝

Contributions are welcome! Please feel free to submit a Pull Request.

License 📄

This project is licensed under the MIT License - see the LICENSE file for details.