SCOPE OF WORK

Lawep.ai required a robust and scalable AI system capable of: Extracting, processing, and organizing an extensive legal dataset. Fine-tuning specialized LLMs tailored for different categories of users. Building a RAG system to ensure responses are contextually grounded and legally accurate. Enabling efficient, reference-backed drafting and research through a web-based interface.

Key Deliverables:

• Deployment of multiple fine-tuned LLMs for diverse legal research needs
• A large-scale text extraction and cleaning system capable of handling hundreds of thousands of documents.
• An intelligent verification system for result accuracy and reference citation.

TECHNOLOGY STACK

• Compute: AWS EC2, SageMaker, AWS Lambda
• Storage: AWS S3
• Database: MySQL, AWS DocumentDB, MongoDB
• Scraping Frameworks: BeautifulSoup (BS4), Puppeteer, JinaAI, Mendable, Scrapegraph.ai
• LLM Hosting: vLLM for high-throughput inference
• LLM Fine-tuning: Huggingface libraries and AWS SageMaker
• Backend Stack: Django (Python)

RESULTS

• High Processing Capacity: Successfully processed over 432,000 legal documents, creating one of the largest specialized datasets for legislative AI work.
• Enhanced Research Speed: Reduced research and drafting time for users by up to 65% compared to manual workflows.
• Reliable Output: The reference extraction system ensures that generated drafts include verifiable, properly cited sources.
• Scalable Architecture: Multiple LLMs deployed across vLLM instances for low-latency, concurrent access.