AI / ML · internship 2025
Roots.
A prototype that recommends crops for a location.
Pick a place, and Roots combines weather data with soil values to suggest the top three crops. Built during a one-month AI/ML internship at ICT Academy of Kerala.
- React
- TypeScript
- Flask
- scikit-learn
- Open-Meteo
fig. 04
soil + climate
- N
- P
- K
- pH
- Rain
top 3 · sample
- Rice71%
- Jute18%
- Coconut11%
What I built
- Random forest classifier trained on 7 features: N, P, K, pH, temperature, humidity and rainfall.
- Flask API that fetches weather from Open-Meteo, builds the feature vector and returns the top 3 crops with probabilities.
- React frontend with location search, browser geolocation and soil sliders.
- Input validation, and a rule-based fallback when the weather API or the model can't answer.
Hard parts
CHALLENGE // 01
Weather APIs only give today's rain
- Problem
- '0 mm right now' says nothing about whether a crop will grow.
- Fix
- Used Open-Meteo's climate archive (1991–2020) to get typical annual rainfall, cached per location.
Limits and next steps
- – Soil values are regional estimates, not lab tests. Results are suggestions, not agronomic advice.