KhetPrithvi

• 

  project interface 2

• 

  project interface 1

India has one of the largest agricultural sectors in the world, yet many farmers struggle with small fragmented farms, lack of accurate crop monitoring, and unpredictable climate conditions. We wanted to build a system that leverages AI + satellite data to solve this problem. Our inspiration came from seeing how geospatial AI can empower farmers with timely and precise crop information, helping improve yields and sustainability.

What it does

KhetPrithvi is an India-first crop classification system that uses IBM’s geospatial and time-series foundation models, fine-tuned for small fragmented farms.

Classifies crops using SAR + optical fusion data

Provides farmers with real-time insights on what crop is being grown where

Lays the groundwork for policy makers, agri-tech startups, and supply chains to make better data-driven decisions

How we built it

Data preprocessing: Collected SAR + optical satellite images, normalized them, and extracted features

Model building: Fine-tuned IBM foundation models for multi-class crop classification

Inference pipeline: Built using PyTorch, FastAPI, and Flask for deployment

Frontend: Simple web interface to upload geospatial data and visualize classification results

Backend: Python + APIs for inference and storage

Challenges we ran into

Handling large geospatial datasets and optimizing preprocessing pipelines

Combining SAR + optical fusion data for better accuracy

Limited compute resources for fine-tuning foundation models

Designing a clean ML pipeline and repo structure that could scale

Accomplishments that we’re proud of

Successfully fine-tuned a state-of-the-art model on fragmented Indian farm data

Deployed a working end-to-end pipeline (data → model → inference → visualization)

Built a clean, well-documented GitHub repo for future improvements

What we learned

Hands-on with IBM geospatial + time-series foundation models

Importance of data preprocessing in remote sensing ML

How to structure ML projects for hackathons + production-readiness

Working as a team under tight deadlines

What’s next for KhetPrithvi

Expanding dataset coverage across more crop types and regions

Building a farmer-facing mobile app with localized language support

Adding weather + soil data integration for richer insights

Collaborating with government and NGOs to reach farmers at scale

Built With

• built-with-languages-&-frameworks:-python
• fastapi
• flask
• hugging-face-tools-&-libraries:-numpy
• ibm-cloud
• javascript-cloud-&-apis:-ibm-foundation-models
• matplotlib-version-control-&-collaboration:-git
• opencv
• pandas
• pytorch
• rasterio
• scikit-learn