Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore how to transform a generalist large language model into a specialized system for interpreting Norwegian building laws and property regulations in this 51-minute conference talk from NDC Oslo. Learn about Norkart's year-long journey developing an LLM-based system that extracts specific information from complex property regulations to help architects and property owners determine building possibilities on specific properties. Discover the challenges faced with initial RAG-system implementations that produced fluctuating and inconsistent results, and understand the strategic decision to pursue deeper customization rather than abandoning the project. Examine the comprehensive series of customizations developed including advanced preprocessing techniques for analyzing and chunking documents while preserving semantic meaning, methods for teaching domain-specific terms and expressions to the system, efficient benchmarking approaches for measuring customized RAG-system performance, and strategies for combining user interface design with AI capabilities to simplify system complexity. Gain insights into the practical challenges of working with intricate legal regulations, the process of taming chaotic legal text with artificial intelligence, and the ultimate success in creating a reliable system capable of answering complex building permission questions.
