AgroLink

What I built

• Built a hybrid ML model forecasting seed demand 6–8 weeks ahead across 60+ districts and 7 agro-ecological zones, blending planting calendars, live weather data and aggregated scan activity.
• Designed a three-layer anti-counterfeiting system: HMAC-SHA256-signed QR codes, per-bag serialisation supporting 99,999+ bags per lot, and GPS fraud mapping of duplicate scans.
• Designed a USSD and SMS fallback modelled on airtime scratch cards, so seed verification and planting guidance reach feature-phone farmers with zero data — no farmer excluded.
• Designed a national seed credit account system — pay at any depot, collect at any depot — with instalment tracking analogous to actuarial cash-flow and credit risk modelling.
• Specified a sub-5KB offline-first PWA landing page loading under 3 seconds on 2G, with Service Worker caching for zero-signal return visits.