Chat with the law

This is a custom implementation of ChatGPT + LangChain + 700+ documents of the law to be accessed as an endpoint API hosted in Azure. Chat experience is optimized for the accuracy of responses at the expense of speed.

Requirements

pip install -r requirements.txt

Overview of important libraries:

• LangChain
• ChatGPT key
• Python 3.8+
• Azure Subscription

Setup

1. Get embeddings through get_embeddings.py and place them in a folder embeddings at the root.
2. Install requirements.
3. Create a .env.prod file with respective keys.
4. Configure your AzureML environment (workspace, compute instances, etc.)
5. Run deployment.py to deploy to AzureML endpoint.

Notes

• Embeddings were created using 512 tokens, with overlapping windows of 40 tokens, and the ADA-002 model from OpenAI.
• Pre-processing steps are kept in a different repository.
• Hybrid search consists of latent-based + keyword-based searches. Results are re-ranked using an out-of-the-box re-ranker (hosted in Azure). Hybrid search is optimized for the most accurate responses at the expense of speed.