I help cities and energy teams turn geospatial data into planning decisions.

Problem

Municipalities need accurate heat-demand maps for decarbonizing space heating, but many lack detailed building-level data.

Method

HeatPrompt uses pretrained vision-language models with an energy-planning prompt to extract semantic features from RGB satellite images, then combines those captions with GIS and building features for heat-demand estimation.

Why it matters

The work shows that visual attributes from satellite imagery can improve heat-demand modeling in data-scarce regions and provide interpretable cues for city-scale building energy planning.

Problem

Building-age distributions are crucial for heat demand mapping, retrofit prioritization, and district energy planning, but local datasets often contain gaps, inconsistencies, and fragmented sources.

Method

A multi-agent LLM system fuses Zensus, OpenStreetMap, and monument data, then uses BuildingAgeCNN with ConvNeXt, FPN, CoordConv, and SE blocks to classify building-age cohorts from satellite imagery.

Why it matters

The pipeline creates structured building-age evidence, provides calibrated confidence estimates, and flags low-confidence predictions for manual review in planning workflows.