Researchers build free, searchable AI database of U.S. local laws

Researchers at the University of California, Berkeley, have launched a groundbreaking, freely accessible database compiling digital ordinances from every U.S. state. Led by technology policy professor Diag Davenport, the initiative addresses the longstanding fragmentation of municipal and county regulations by aggregating nearly seven million pages of local laws from approximately 9,000 jurisdictions into a single, machine-readable repository. Historically, accessing and comparing local rules has required navigating inconsistent government websites and dense, unstructured PDFs, creating significant barriers for developers, residents, and researchers. To overcome these challenges, the Berkeley team deployed a multi-stage artificial intelligence pipeline. They began by systematically harvesting documents from municipal portals and third-party archives, adhering to site-specific technical constraints. The team then utilized a vision-language optical character recognition model, LightOnOCR, to extract and standardize text and structural data from poorly formatted scans. Subsequent processing by large language models from OpenAI enabled automated tagging and categorization, transforming unstructured ordinances into a searchable corpus. The publicly released database, launched in June, fundamentally alters how communities and regulators can be analyzed. By standardizing municipal codes, the project enables unprecedented cross-jurisdictional comparisons of regulations governing housing, construction, public spaces, and everyday conduct. The structured data also serves as a foundation for third-party applications, including specialized AI chatbots and custom regulatory databases that can instantly contrast local requirements. For real estate developers, these tools could streamline compliance and cost estimation. For journalists and public policy analysts, the repository eliminates hours of manual research, enabling large-scale studies on regulatory clarity, enforcement patterns, and systemic biases across historically disparate jurisdictions. Academic recognition of the project arrived when the research paper detailing the methodology and dataset was accepted for presentation at the Conference on Neural Information Processing Systems in December. Moving forward, the Berkeley team emphasizes that all derivative public tools should remain free and openly accessible. By rendering local governance transparent and computationally tractable, the initiative aims to empower citizens and officials alike to evaluate which regulations function effectively, who they serve, and how municipal policy can be optimized for modern societal needs.