GENIUS: GENerative Intelligence for Urban Sustainability
2025 - Present
SIERA
Project Leads
PI: Norhan Bayomi,
MIT SIERA, Research Scientist
PI: Omar Khattab,
MIT EECS, Assistant Professor
PI: John E. Fernandez,
MIT UMERA, Faculty Director
Project Team
Ziyad Hassan
Graduate Research Assistant - MIT EECS
Nathanial Morgan
Undergraduate Research Assistant - MIT EECS
Audrey Lin
Undergraduate Research Assistant - MIT EECS
Supported By
MIT Generative AI Impact Consortium
An AI-powered analytical platform that harnesses generative AI and retrieval-augmented generation to convert unstructured urban climate policies from 100 diverse cities globally into structured, machine-readable datasets, enabling rigorous cross-city comparison and evidence-based policy transfer through the Scities platform.
GENIUS (GENerative Intelligence for Urban Sustainability) is an analytical infrastructure developed by SIERA lab in collaboration with MIT EEC, Omar Khattab, to address a critical gap in urban climate governance: the absence of a standardized, scalable framework for systematically assessing and comparing climate policies across cities. Cities contribute roughly 70% of global CO₂ emissions, yet the thousands of policy documents produced each year remain fragmented and difficult to analyze at scale using conventional methods.
The platform harnesses generative AI, domain-specific prompt engineering, and retrieval-augmented generation (RAG) pipelines to transform raw, unstructured urban policy documents into structured, machine-readable data. Policies are classified according to standardized frameworks drawn from IPCC reporting guidelines and modern urban sustainability protocols, capturing key elements such as targets, implementation mechanisms, timelines, and monitoring approaches. This structured representation ensures consistency and comparability across diverse jurisdictions and policy formats.
At the core of GENIUS is the integration of city typology data from the Scities platform with policy content, enabling contextually grounded cross-city comparisons. Rather than treating all cities as equivalent, the system leverages the UMERA’s classification of 11,000 cities globally, through the lens of resource use, climate risks, and environmental impacts, to identify structurally and metabolically comparable urban environments. This supports knowledge transfer between cities facing analogous climate challenges, helping policymakers identify and adapt successful strategies rather than starting from scratch.
The project follows a four-phase methodology: dataset curation across 100 globally diverse cities with annotation of approximately 20% of the corpus for validation; generative AI training and optimization using retrieval-augmented generation; development of a cross-policy analysis framework with novel visualization techniques; and prototype testing with 15–20 urban planners and researchers.