AgriPulse

Problem

Smallholder farmers in Malawi (who produce 80% of the country's food) make planting and input decisions with almost no data. Poor decisions cost them 20–40% of potential yield annually.

Solution

AgriPulse combines historical climate data, soil classification maps, and FAOSTAT crop yield records to train a predictive model per district and crop type. Extension workers use a Streamlit dashboard to input local conditions and receive yield forecasts with confidence intervals and planting recommendations.

Real-World Impact

Piloted in the Northern Region. Extension workers reported the tool reduced their advisory time by 30% and gave farmers a concrete, data-backed forecast to plan around.

Challenges Faced

Missing and inconsistent historical data for Malawi required significant imputation and cross-validation against neighbouring countries' datasets. Spatial data alignment was particularly complex.

Key Learnings

Real-world agricultural ML requires domain expertise as much as technical skill. Working with an agronomy student to validate model outputs was crucial.

Demo & Execution Screenshots