Legal Brief Companion - RAG-Powered Legal and Technical Assistant

This project implements a modular Retrieval-Augmented Generation (RAG) assistant using LangChain, designed to answer legal and technical queries based on custom documents. It features document ingestion, semantic retrieval, prompt templating, and LLM-backed response generation—all wrapped in a clean CLI and Streamlit interface.

Project Structure

legal_brief_companion/
├── src/
│   └── legal_brief_companion/
│       ├── __init__.py
│       ├── config/
│       │   └── settings.py           # Application configuration and env loading
│       ├── ingestion/
│       │   ├── document_loader.py    # Load PDFs/texts into the pipeline
│       │   └── text_splitter.py      # Chunk large documents
│       ├── retrieval/
│       │   ├── vector_store.py       # Manage vector store persistence (Chroma)
│       │   └── retriever.py          # Query and rank relevant chunks
│       ├── llm/
│       │   ├── chain.py              # LLM orchestration and chains
│       │   └── prompt_templates.py   # Reusable prompt templates
│       ├── interface/
│       │   └── cli.py                # CLI for local usage
│       ├── utils/
│       │   └── helpers.py            # Shared utilities and helpers
│       └── tests/                        # Unit tests
│       |     ├── test_chain.py
│       |     ├── test_ingestion.py
│       |     └── test_llm.py
│       └── ingest.py                 # CLI/script entry for ingestion
├── data/
│   ├── documents/                    # User documents (PDFs, txt)
│   └── vector_store/                 # Persisted vector DB files
│       └── <instance-id>/
|
├── app.py                            # Streamlit / main app entry
├── pyproject.toml
├── requirements.txt
├── .env
├── .gitignore
└── README.md

🚀 Quick Start

Prerequisites

• Python 3.9+
• Groq API key
• pip or poetry

⚙️ Setup Instructions

1. Clone the repository:

   git clone <your_repository_url>
   cd legal_brief_companion

2. Install the required dependencies:

   pip install -r requirements.txt

3. Set up your environment variables in the .env file. Important: Do not hardcode your API key directly in the .env file. Instead, set it as an environment variable in your system.

   GROQ_API_KEY=your_groq_key
   EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
   LLM_PROVIDER=groq
   LLM_MODEL=llama3-8b-8192
   DATABASE_URL=sqlite:///data/knowledge_base.db
   DOCUMENTS_PATH=data/documents
   PERSIST_DIRECTORY=data/vector_store
   MODEL_NAME=llama3-8b-8192
   VECTOR_STORE_TYPE=chroma
   
   debug = true
   

4. Set your Groq API key as an environment variable:

   # PowerShell
   $env:GROQ_API_KEY="your_api_key_here"
   
   # Linux/macOS
   export GROQ_API_KEY="your_api_key_here"

5. Ingest your documents: This process will create the vector store.

   python src/legal_brief_companion/ingest.py

6. Run the application:

   streamlit run app.py

🧑‍⚖️ Usage

After running the application, you can interact with the assistant through the command-line interface or the Streamlit interface. Provide your queries, and the assistant will respond based on the ingested documents.

🧩 Components

• Document Ingestion: Load and chunk legal PDFs or text files using document_loader.py and text_splitter.py.
• Vector Store Retrieval: Embed and retrieve relevant chunks using ChromaDB, managed by vector_store.py and retriever.py.
• LLM Interaction: Generate answers using Groq-hosted LLaMA models, orchestrated by chain.py.
• Prompt Templates: Modular prompts for transparent reasoning, defined in prompt_templates.py.
• User Interface: Streamlit frontend (app.py) and CLI (cli.py) for flexible interaction.
• Config Management: Pydantic-based .env loader for reproducibility via settings.py.

🧩 Core Components

Document Processing

• PDF and text file ingestion
• Intelligent document chunking
• Semantic embedding generation

Retrieval System

• ChromaDB vector store integration
• Similarity-based chunk retrieval
• Context ranking and selection

LLM Integration

• Groq-hosted LLaMA models
• Structured prompt engineering
• Response chain orchestration

🧪 Testing

Basic unit tests are included under src/legal_brief_companion/tests/. Run with:

pytest src/legal_brief_companion/tests/

🛠️ Built With

• LangChain
• ChromaDB
• Groq
• Streamlit
• Pydantic

🔒 Security Notes

• Never commit your .env file or API keys to GitHub. The .gitignore file is set up to exclude secrets and unnecessary files.
• Use environment variables for secrets
• Follow security best practices
• Regular dependency updates

🔒 License

This project is licensed under the MIT License.

📬 Contact

Built by Getahune Wondemenhu Alemayhu.

• Email: [getahune.alemayhu@gmail.com] For questions, contributions, or collaboration, feel free to reach out or open an issue.

Contributions are welcome!

Please open issues or submit pull requests for improvements or bug fixes. Follow the existing code style and include tests for new features and suggestions.