Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/pratham-jaiswal/legal-ai
Legal QA system using LangChain and Ollama to provide legal advice based on Indian legal documents.
https://github.com/pratham-jaiswal/legal-ai
cohere flask langchain langchain-python llm ollama retrieval-augmented-generation
Last synced: 29 days ago
JSON representation
Legal QA system using LangChain and Ollama to provide legal advice based on Indian legal documents.
- Host: GitHub
- URL: https://github.com/pratham-jaiswal/legal-ai
- Owner: pratham-jaiswal
- License: mit
- Created: 2024-09-11T12:55:31.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2024-11-10T08:43:11.000Z (2 months ago)
- Last Synced: 2024-12-18T19:12:08.263Z (about 1 month ago)
- Topics: cohere, flask, langchain, langchain-python, llm, ollama, retrieval-augmented-generation
- Language: Jupyter Notebook
- Homepage: https://legal-ai-chatbot.streamlit.app
- Size: 31 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
- License: LICENSE
Awesome Lists containing this project
README
# Legal AI - Legal QA System
This is a legal question-answering system that utilizes LangChain to process and query legal documents. The system leverages Large Language Models, Natural Language Processing and Embeddings to provide accurate legal advice based on the provided documents.
> [!NOTE]
> The retrieval-augmented generation (RAG) model utilized is tailored to the Indian legal context.
***Visit the [flask](https://github.com/pratham-jaiswal/legal-ai/tree/flask) branch for the flask app.***
## Context Documents Used:
- The Bharatiya Nyaya Sanhita, 2023 - *./context/250883_english_01042024.pdf*
- The Constitution of India, 2024 - *./context/20240716890312078.pdf*
- Indian Polity by M. Laxmikanth 7th Edition
> [!NOTE]
> The document "Indian Polity by M. Laxmikanth, 7th Edition" is referenced for context but is not included in this repository. To recreate the Embedding Vector DB and use this file, you need to acquire it separately and place it in the `./context/` directory. Alternatively, you can use the provided `./chroma_db/`, which includes this document.
## Getting Started
1. Clone the repository:
```sh
git clone https://github.com/pratham-jaiswal/legal-ai.git
```
2. Install the required libraries:
```sh
pip install -r requirements.txt
```
3. **Choose your LLM setup**:
- **Local LLM (Ollama)**: Download and install [Ollama](https://ollama.com/download)
- **LLM Providers (OpenAI, Cohere, Anthropic, etc.)**: Obtain the API key from your preferred provider:
- [OpenAI](https://platform.openai.com/api-keys)
- [Cohere](https://dashboard.cohere.com/api-keys)
- [Anthropic](https://console.anthropic.com/settings/keys)
- [Any other supported ones](https://python.langchain.com/docs/integrations/chat/)
Create a `.env` file to store the API key as shown below. You can skip steps **4** and **5** if using an API key:
```env
COHERE_API_KEY=Your-api-key
```
> [!NOTE]
> LLM providers may charge a fee. Cohere offers a free "Trial Key".
4. Start Ollama (*Not needed when using a LLM provider*)
```sh
ollama serve
```
5. Pull the required model(s) from Ollama (e.g., `phi3.5`). You can explore other models from the [Ollama library](https://ollama.com/library). (*Not needed when using a LLM provider*)
```sh
ollama pull phi3.5
```
6. Run the `main.ipynb` file to start your project.
## Changes you can make
1. Use a different LLM model. Ollama offers [multiple models](https://ollama.com/library) which you can pull and use. For example
```sh
ollama pull llama3.1:8b
```
```py
llm = Ollama(model="llama3.1:8b")
```
You can also use OpenAI, Cohere, Anthropic, or any other LLM - Ollama is just free and can be pulled to be used locally.
```py
llm = ChatCohere(model="command-r")
```
> [!NOTE]
> You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models. [Source](https://github.com/ollama/ollama?tab=readme-ov-file#model-library).
> Phi3.5 is a 3B model, and I'm on a 8GB RAM i5 9th gen 4 year old laptop, and it takes about 10 mintues to create a vector db with around 4.8 chunks of context data, and 2-5 mins to answer a query.
2. Use different chunking methods [1](https://python.langchain.com/v0.2/docs/how_to/#text-splitters) [2](https://blog.lancedb.com/chunking-techniques-with-langchain-and-llamaindex/), or parameters. For example,
```py
text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=70)
```
> [!NOTE]
> The smaller the chunk_size, the more number of chunks which means it will take more time to complete the process.
3. Use different embedding models provided by [sentence-transformers](https://huggingface.co/sentence-transformers#models), [Ollama](https://ollama.com/blog/embedding-models), OpenAI, and more. For example,
```py
embeddings = SentenceTransformerEmbeddings(model_name="all-mpnet-base-v2")
```
> [!NOTE]
> The size of embedding model, and your hardware specification will affect the time it will take to create the Vector DB.
4. Use a different Vector Store such as [Chroma](https://python.langchain.com/v0.2/docs/integrations/vectorstores/chroma/), [Faiss](https://python.langchain.com/v0.2/docs/integrations/vectorstores/faiss/), and [more](https://python.langchain.com/v0.2/docs/integrations/vectorstores/).
5. Manipulate the prompt to refine or change the responses.
6. Replace the documents in `./context/` with those relevant to your legal department or country.
## Contributing
Feel free to submit issues or pull requests if you have suggestions or improvements.
## License
This project is licensed under the MIT License - see the [LICENSE.md](https://github.com/pratham-jaiswal/legal-ai/blob/main/LICENSE) file for details.