GeoContextualize

GeoContextualize was inspired by the challenge of turning raw satellite data into actionable insights. While global Earth observation datasets are abundant, extracting meaningful, location-specific context remains complex and time-consuming for non-experts.

This project automates that process. Users submit an area of interest, and the system retrieves relevant satellite data—digital elevation models, vegetation indices (NDVI), and land-cover maps—via the Microsoft Planetary Computer. Raster data is clipped, summarized, and semantically labeled, producing accurate quantitative insights. An AI model then generates audience-specific narratives (academic, policy, investor, or practitioner) based strictly on the computed data.

Building this project taught me the importance of separating data computation from AI interpretation to maintain reliability and avoid hallucinations. I also gained experience deploying geospatial workloads in cloud environments, handling API reliability, and optimizing large raster processing.

The main challenges were managing memory and request limits, ensuring real-time performance, and designing prompts that keep AI outputs faithful to the data.

The result is a lightweight, intuitive tool that transforms complex satellite imagery into human-readable insights, useful for research, planning, agriculture, and climate analysis.

Built With

• fastapi
• python