Creating an AI-Powered Legal Assistant for Swiss Law Professionals
Our advanced solution leverages GenAI, RAG technology, and an agent-based LLM system to process over 100,000 legal documents, delivering accurate insights and efficient information extraction from massive legal datasets.
Law professionals have to process large volumes of legal documentation and cases to deliver their services as data extraction and analysis are time-consuming and resource-intensive. Based on the user’s well-formulated prompts, RNDpoint’s AI assistant automates legal data processing and enables intelligent, accurate insights from extensive legal documentation datasets.
Client’s Profile
The Client is an individual entrepreneur currently occupying the position of AI lead at a tech giant. The Client’s business idea was to create an AI-powered legal assistant for Swiss legal professionals. The assistant had to be trained on Swiss legal data to aid legal experts in locating relevant cases and streamlining their legal analysis activities.
Client’s Requirements/Goals
The Client provided us with the following technical requirements for the AI legal assistant project:
- Advanced GenAI platform development. The legal assistance platform had to enable simple and user-friendly navigation of relevant legal information and its extraction from 100,000+ legal documents (court rulings, Swiss laws, and academic articles).
- Information retrieval. The platform’s users had to receive relevant sources and quick summaries in response to their prompts or get accurate answers to questions. The retrieval accuracy had to be enhanced by implementing Retrieval-Augmented Generation (RAG) technology.
- Agent-based LLM system. The platform had to utilize the combination of CrewAI and LangGraph technologies in its LLM system.
- Data integration capabilities. The AI assistant had to be able to extract data from the custom-built Azure CosmosDB and open web sources to respond to user inquiries.
- Rigorous testing. The platform had to go through LangSmith testing before deployment.
Business Challenge
The RNDpoint team encountered several challenges in the process of working on the AI legal assistant.
- Big data. The project involved processing over 100,000 legal documents to create the database for AI legal assistant training and data retrieval, which is demanding in terms of data handling and storage.
- Complexities of legal jargon. It is hard for legal non-professionals to navigate the complex legal terminology and concepts, which is still vital for accurate information categorization and further retrieval.
- Technology stack. Orchestrating various technologies, from GPT models to OpenAI embeddings and LangChain, during their integration within the AI legal platform.
- Real-time response modeling. The AI legal assistant had to provide responses and quick summaries in real-time, which requires advanced computing capacity.
- Data extraction and homogenization. The assistant had to extract data fragments from various legal sources and turn them into a cohesive response before submitting the answer to the user.
Implementation
The RNDpoint team implemented the AI legal assistant across several technical environments and technologies.
Agent-Based LLM System Design
The agent-based LLM system was created using CrewAI and LangGraph technologies. The RNDpoint team utilized OpenAI’s GPT-4 and GPT-4o APIs to enable human language understanding and generative content creation.
Data Integration Activities
Azure CosmosDB was created from scratch to include 100,000+ legal documents covering Switzerland’s extensive legal documentation. The database included legal case proceedings, laws, and articles in legal science to train the AL model for accurate, intelligent legal prompt analysis. The RNDpoint architects also equipped the AI legal assistant with an API for data search in open web sources to ensure the analysis stays up to date with the latest legal changes and publications.
RAG Implementation
We implemented RAG using custom Azure Functions to guarantee a high level of information retrieval accuracy. The Open Embeddings technology was coupled with WeaviateDB ensuring efficient vector search and similarity matching within user prompt processing.
Platform Development
The team chose Streamlit for UI development to create a user-friendly, interactive interface. Conversational flow management and interactions were technically enabled with a combination of LangChain and CrewAI/LangGraph tools.
Audit and Evaluation
The platform’s compliance with the intended performance goals and reliability standards was checked with the help of LangSmith technology.
Solution Delivered
The resulting AI-powered software product delivered by our team represents a robust, intelligent, and comprehensive legal information platform with the following characteristics:
- GenAI assistant navigating an internal database of 100,000+ legal documents and retrieving data from open web sources in real-time.
- Delivery of accurate, relevant responses, data summaries, and precise answers to user prompts.
- Efficient data handling with the combination of CosmosDB and open web source access.
- Effective and safe data storage and retrieval mechanisms powered by WeaviateDB and RAG.
- High-quality NLP technology using several advanced GPT-4 models.
- Agent-based LLM system capable of intelligent query handling and response generation.
Project Outcomes
The project was initiated in January 2024 and is still under development. At present, the team working on the AI platform includes two AI engineers and one front-end engineer. As soon as the AI platform gets launched, it is expected to bring essential value to Swiss legal professionals by enabling accurate data processing and automation of time-consuming legal analysis and data retrieval processes in a natural conversion-like manner.