LawBot: Enhancing LLMs with RAG for Legal Precision

Group Project of COMP0087: Statistical Natural Language Processing

Introduction

LawBot is a simple yet effective framework tailored to specialized legal domains, operational without training. Utilizing Chinese legal and regulatory documents as the knowledge base, LawBot enhances the breadth of retrieval through multi-query generation and hybrid search strategies. It increases precision with metadata filtering and confirms the plausibility of knowledge through context-based reranking. Remarkably, all these procedures are conducted via zero-shot prompting, making LawBot broadly applicable even when LLMs are accessible only through a black-box API.

Installation

Follow these steps to set up the LawBot environment on your local machine:

1. Clone the repository:

   git clone https://github.com/yix8/LawBot.git

2. Navigate to the LawBot directory:

   cd LawBot

3. Install the required packages:

   pip install -r requirements.txt

4. Build the general vector store:

   python framework/embed_laws.py

5. Build the specific vector store:

   python framework/finetune_data/embed_query.py

Configuration

Set up the necessary API keys in the .env file located in the framework folder:

• OPENAI_API_KEY: Your OpenAI API key.
• COHERE_RERANK_KEY: Your Cohere rerank API key.
• LANGCHAIN_API_KEY: Your LangChain API key.
• LANGCHAIN_PROJECT: Your LangChain project identifier.

Running LawBot

To interact with the model via a web interface:

python framework/App.py

China Law Query Synthetic

We also proposed an open-source QA dataset, the Chinese Legal Question Answering dataset (CLQS) which can be utilized as an instruction dataset.