Projects
Urban Geo Analytics contributes to international research and consulting projects at the intersection of geospatial AI, street-level imagery analysis, and urban planning. Below are the projects the firm is or has participated in, from European research consortia to client assignments.
AI for Last-Mile Analysis in West Bengal
The World Bank
Running: 2026
Slow road movement in Indian cities is not only a peak-congestion problem. It reflects continuous competition for limited road space among vehicles, pedestrians, vendors, and other uses, creating persistent frictions that are especially important for freight in the first and last mile. While truck GPS data can reveal where speed reductions occur, it cannot explain their causes. Street-level imagery provides direct visual evidence, and Vision-Language Models offer a practical way to convert this imagery into structured, planner-ready indicators at scale. The assignment with the World Bank establishes a rigorous and reproducible benchmark for this approach. It includes the development of a freight-friction indicator framework, a quality-controlled benchmark dataset with expert annotations, the evaluation of multiple open-source Vision-Language Models, and fully documented, locally executable benchmarking pipelines. The objective is to validate the methodology before large-scale inference and mapping in subsequent phases. UGA is the sole consultant for the assignment, leading the design of the indicator framework, dataset curation and annotation, model evaluation, pipeline development, and technical reporting.
Client: The World Bank (Transport / Logistics), Washington, DC.
Study area: West Bengal, India.
EMC2 — The Evolutive Meshed Compact City
Horizon Europe — Driving Urban Transitions (DUT) Partnership, 15-minute City Transition Pathway
Running: 2023 – 2026
The 15-minute city has demanding prerequisites in urban form, and the transition is hardest in suburbs and car-dependent peripheries. EMC2 proposes a new model adapted to these contexts: compact urban development organised as corridors along existing main roads, forming a meshed structure across the metropolitan area. Redesigned for pedestrians, these interconnected roads become vibrant, inclusive main streets offering a high quality of stay, a wide variety of mixed uses, and connections to wider-range mobility options, while the suburban fabric within the mesh requires only marginal improvement. The model is assessed in five European case studies (Nice, Lille–Roubaix–Tourcoing, Vienna, Gothenburg, and the Versilia conurbation) through a triangulated methodology combining geospatial and network modelling, morphometrics, observational usage analysis and planning-oriented evaluation. Within EMC2, UGA designed and developed the project’s AI streetscape-analysis tooling under contract with CNRS: SAGAI, the open-source workflow for VLM-based scoring and mapping of street-level urban scenes, and UVLM, the unified multi-model VLM inference and benchmarking framework.
Consortium: CNRS – Université Côte d’Azur, Agence d’Urbanisme Azuréenne and Agence de Développement et d’Urbanisme de Lille Métropole (France), Chalmers University of Technology – SMoG and the City of Gothenburg (Sweden), University of Pisa – DESTeC and the Cities of Viareggio and Massa (Italy), and TU Wien – Department of Urban Design (Austria).
e-Geopolis — Harmonized Measurement of World Urbanisation
e-Geopolis Institute — long-term scientific partnership
Running: ongoing
Official urban statistics are not comparable across borders: every country defines “urban” differently, making global claims about urbanisation difficult to verify. The e-Geopolis programme addresses this by applying a single, harmonized definition worldwide — an agglomeration is a continuously built-up area of at least 10,000 inhabitants — and by combining census records, satellite and aerial imagery, and cartographic sources to map and document the population trajectories of tens of thousands of agglomerations over the long term. The programme’s regional databases, including Africapolis (produced with the OECD Sahel and West Africa Club and covering more than 7,500 agglomerations across the continent) and Indiapolis (developed with the French Institute of Pondicherry), have become reference datasets for research and development policy, feeding into work on the New Urban Agenda and SDG 11. Urban Geo Analytics is a long-term participant in the e-Geopolis Institute’s work on world population and urbanisation, contributing geospatial processing, automation, and AI-assisted workflows for the delineation of built-up areas, the integration of heterogeneous demographic sources, and the updating of agglomeration databases across world regions.
AI for Last-Mile Analysis in West Bengal
The World Bank
Running: 2026
Slow road movement in Indian cities is not only a peak-congestion problem. It reflects continuous competition for limited road space among vehicles, pedestrians, vendors, and other uses, creating persistent frictions that are especially important for freight in the first and last mile. While truck GPS data can reveal where speed reductions occur, it cannot explain their causes. Street-level imagery provides direct visual evidence, and Vision-Language Models offer a practical way to convert this imagery into structured, planner-ready indicators at scale. The assignment with the World Bank establishes a rigorous and reproducible benchmark for this approach. It includes the development of a freight-friction indicator framework, a quality-controlled benchmark dataset with expert annotations, the evaluation of multiple open-source Vision-Language Models, and fully documented, locally executable benchmarking pipelines. The objective is to validate the methodology before large-scale inference and mapping in subsequent phases. UGA is the sole consultant for the assignment, leading the design of the indicator framework, dataset curation and annotation, model evaluation, pipeline development, and technical reporting.
Client: The World Bank (Transport / Logistics), Washington, DC.
Study area: West Bengal, India.
EMC2 — The Evolutive Meshed Compact City
Horizon Europe — Driving Urban Transitions (DUT) Partnership, 15-minute City Transition Pathway
Running: 2023 – 2026
The 15-minute city has demanding prerequisites in urban form, and the transition is hardest in suburbs and car-dependent peripheries. EMC2 proposes a new model adapted to these contexts: compact urban development organised as corridors along existing main roads, forming a meshed structure across the metropolitan area. Redesigned for pedestrians, these interconnected roads become vibrant, inclusive main streets offering a high quality of stay, a wide variety of mixed uses, and connections to wider-range mobility options, while the suburban fabric within the mesh requires only marginal improvement. The model is assessed in five European case studies (Nice, Lille–Roubaix–Tourcoing, Vienna, Gothenburg, and the Versilia conurbation) through a triangulated methodology combining geospatial and network modelling, morphometrics, observational usage analysis and planning-oriented evaluation. Within EMC2, UGA designed and developed the project’s AI streetscape-analysis tooling under contract with CNRS: SAGAI, the open-source workflow for VLM-based scoring and mapping of street-level urban scenes, and UVLM, the unified multi-model VLM inference and benchmarking framework.
Consortium: CNRS – Université Côte d’Azur, Agence d’Urbanisme Azuréenne and Agence de Développement et d’Urbanisme de Lille Métropole (France), Chalmers University of Technology – SMoG and the City of Gothenburg (Sweden), University of Pisa – DESTeC and the Cities of Viareggio and Massa (Italy), and TU Wien – Department of Urban Design (Austria).
e-Geopolis — Harmonized Measurement of World Urbanisation
e-Geopolis Institute — long-term scientific partnership
Running: ongoing
Official urban statistics are not comparable across borders: every country defines “urban” differently, making global claims about urbanisation difficult to verify. The e-Geopolis programme addresses this by applying a single, harmonized definition worldwide — an agglomeration is a continuously built-up area of at least 10,000 inhabitants — and by combining census records, satellite and aerial imagery, and cartographic sources to map and document the population trajectories of tens of thousands of agglomerations over the long term. The programme’s regional databases, including Africapolis (produced with the OECD Sahel and West Africa Club and covering more than 7,500 agglomerations across the continent) and Indiapolis (developed with the French Institute of Pondicherry), have become reference datasets for research and development policy, feeding into work on the New Urban Agenda and SDG 11. Urban Geo Analytics is a long-term participant in the e-Geopolis Institute’s work on world population and urbanisation, contributing geospatial processing, automation, and AI-assisted workflows for the delineation of built-up areas, the integration of heterogeneous demographic sources, and the updating of agglomeration databases across world regions.