DEV Community

joseph king'ori
joseph king'ori

Posted on

Roast My Stack: AI-Powered USSD Marketplace for 7M Kenyan Farmers

I'm building KamataShamba – an AI agricultural platform accessible via USSD on any basic phone.

The Problem
Kenyan smallholder farmers:

Lose 30-50% of potential income

Have no weather/pest alerts

Sell to middlemen at 50% below market

Can't access credit (no formal history)

60%+ use basic phones only

The Solution
Feature How Channel
Weather forecast AI + satellite USSD
Pest detection Image recognition WhatsApp
Marketplace Farmer → buyer direct USSD + Web
Credit scoring Platform activity Automatic
Loans/Insurance Partner integration USSD

The Stack
yaml
Channels:
USSD: Africa's Talking API
WhatsApp: Twilio/Meta
Mobile (future): Flutter
Web (buyers/admin): React

Backend:
API: Python/FastAPI
Cache/Queue: Redis
Primary DB: PostgreSQL
Time-series: TimescaleDB

ML:
Pest detection: TensorFlow (CNN)
Weather forecast: TensorFlow (LSTM)
Credit scoring: Scikit-learn

Infrastructure:
Cloud: AWS (EKS, RDS, S3)
Container: Docker

Payments:
M-Pesa API (C2B/B2C + escrow)
Questions for the Community
USSD scaling: Africa's Talking docs mention ~100 concurrent sessions. How do I handle 10,000+ farmers dialing in simultaneously during harvest?

Pest detection training: Public datasets (PlantVillage) have limited African crops. Should I partner with KALRO for labeled local data?

Matching engine (farmer → buyer): Need real-time matching by location, product, quantity, price. Redis sorted sets? Postgres LISTEN/NOTIFY? Kafka?

Offline challenges: Network in rural Kenya is unreliable. How should I design for offline-first?

Government APIs: Anyone worked with Kenya Space Agency or KAMIS APIs? Reliable?

This is open-source friendly. DM me if you want to contribute or collaborate

Top comments (0)