Case study Geoverdant
Satellite data everyone had, and nobody could use
Farmers, environmental consultants, and site planners all share one blind spot: land changes gradually, and by the time you can see it, the cheap fix has passed. The data to catch it early is free and public. Reading it required a GIS specialist. We closed that gap.
Services Platform architecture, product design & engineering
Category GeoTech / data intelligence platform
Client GeoVerdant
Stack Next.js · MapLibre GL · NestJS · FastAPI · PostgreSQL · rio-tiler · Redis
The problem
They had the data, but no clarity.
Public satellite imagery is abundant. Turning it into a decision took either a consultant visit or a desktop GIS tool built for specialists.
✓No simple way to track land health over time ✓Historical environmental context effectively inaccessible ✓Satellite tooling too technical for the people deciding ✓Site visits too slow and too expensive to repeat ✓Decisions made on assumptions and outdated reports ✓Problem areas invisible until they were obvious
What it was costing
✓Farmers planning crops without an early read on stress ✓Developers assessing sites on incomplete information ✓Environmental trends going unnoticed until irreversible ✓Every decision reactive instead of planned
Our approach
Draw a boundary, get an analysis
Three independently deployable services: a dashboard for people, an API that owns accounts and access, and a Python pipeline that does the science. Separated on purpose, so the user-facing parts stay fast while the heavy processing scales on its own.
✓Interactive area-of-interest drawing on a MapLibre map ✓NASA HLS scene retrieval with cloud and shadow masking ✓Six vegetation and moisture indices computed per pixel ✓Areas over 500 km² split into sub-regions and merged automatically ✓Statistical rigour: Mann-Kendall trend tests, Sen's slope, z-score anomalies ✓Results served as real map tiles, so users pan and zoom like a basemap
Core capabilities
What it does now
Raw satellite pixels turned into a decision a non-specialist can act on.
✓NDVI, EVI, NDWI, SAVI, NDRE, and LAI from one request ✓Time-series trends with statistically detected anomalies ✓Problem zones returned as geometry, not just a number on a chart ✓HMAC-signed tile URLs for high-throughput map rendering ✓Durable background jobs with retries and checkpoints ✓An MCP server, so AI assistants can run the workflow too