Skip to content
I.STEM Lab

Research in GEO-AI & spatial intelligence

Our research is applied: every study is aimed at a tool someone will use to make a better decision.

Satellite imagery super-resolution research

Papers & studies

  • 2025

    Advancing satellite imagery super-resolution for agricultural monitoring

    Deep learning methods that enhance low-resolution satellite images so small fields, roads and settlements can be mapped reliably.

    Ongoing research

    GeoAI · Remote sensing

  • 2025

    Integrated food security monitoring with Earth observation data in Karnali Province

    Combining remote sensing indicators with regional data to support food security planning and early warning.

    Applied study

    GeoAI · Food security

  • 2026

    Grounded question answering over geospatial reports using RAG

    Evaluating retrieval-augmented generation for answering planners' questions from technical and policy documents with citations.

    Working paper

    AI Systems · RAG

Research areas

What we study, and why

We research problems that block real projects. A study is worth doing when its result changes a decision, a tool or a method that people in Nepal and similar contexts actually use.

GeoAI and remote sensing
Machine learning on satellite, aerial and drone imagery: super-resolution, land cover and change detection where ground data is scarce.
Food security and agriculture
Turning imagery, weather and survey data into early, explainable signals about crops and vulnerable districts.
RAG for technical documents
Retrieval-augmented assistants that answer from reports and policy documents, with sources, and how to test that they are right.
Reality capture and immersive 3D
Photogrammetry, laser scanning and Gaussian splatting to record places, and how to publish them for the web and headsets.

How we work

Applied, measured and shared

  • We start from a decision someone needs to make, not from a technique.
  • We test on data held back from training and report accuracy, error and uncertainty honestly.
  • We prefer free and open data, such as Sentinel-2 and Landsat, so results can be repeated.
  • We publish what we learn as plain-language material on our Learn pages.

Read the background

Start a project

Partner with us on research

We collaborate with universities, research institutes and agencies on GeoAI studies, grant proposals and pilots.