Krishak AI.
An AI-powered mobile assistant for farmers.
The project
& the challenge.
Krishak AI puts an agronomist in every farmer's pocket. The app uses image recognition and LLM-driven Q&A to identify crop diseases, recommend treatments, and guide farmers through seasonal decisions in their own language.
Smallholder farmers often lack access to extension officers or reliable agronomic advice, leading to crop loss from preventable disease and poor input timing. Existing apps assumed high literacy, English fluency, and stable connectivity — none of which is realistic in rural Bangladesh.
How I built it.
Step-by-step decisions and trade-offs that shaped the final shipped product.
- 01
Built the mobile app in Flutter with a clean Bangla-first UI, voice input, and image capture flow optimized for low-end Android devices.
- 02
Engineered a Node.js + Express.js backend that orchestrates calls to vision models for disease detection and an LLM for follow-up questions and treatment guidance.
- 03
Stored user farms, crop history, and consultation logs in MongoDB so the assistant could give context-aware advice over time.
- 04
Built an offline-first caching layer so common diagnoses and seasonal tips work even on intermittent connectivity.
- 05
Added a feedback loop where farmer outcomes flow back into model fine-tuning datasets.
What it does.
Crop disease detection
Snap a photo of a leaf — get an instant diagnosis and treatment plan.
AI agronomist chat
Ask questions in Bangla; get LLM-backed advice tailored to your farm.
Field intelligence
Track multiple plots, growth stages, and per-crop recommendations.
Offline-first
Works on patchy connectivity with smart caching and sync.
Voice input
Speak instead of typing — designed for low-literacy users.
Bangla-first UX
Native language interface, units, and cultural context throughout.
