MassivCart: Building a Price Map of Jamaica in a Weekend

In Jamaica, the price of a pound of flour can swing 40% between a supermarket in Half-Way-Tree and one ten minutes away in Liguanea. At the Intellibus Hackathon this March, I built Massiv Cart to fix that: a crowdsourced grocery price-intelligence platform where shoppers snap a photo of their receipt on Telegram, and the system extracts, structures, and publishes the prices back to the community. One photo per shop, thousands of shoppers, and suddenly you have a living price map of the country.

The stack is deliberately light. Next.js handles the interface with no app install, no onboarding friction. Claude Vision reads the receipts, pulling out line items and prices with surprising accuracy on crumpled, glare-hit thermal paper. Supabase stores the normalized data, Kafka handles the stream of incoming photos so nothing gets dropped during a burst, and everything runs on GCP Cloud Run to keep idle costs near zero. The architecture isn't clever for clever's sake; it's the shortest path from "shopper leaves the store" to "someone else saves money on the same basket tomorrow."

MassivCart didn't place at the hackathon, but the exercise was worth more than the placement. Building it end-to-end in a weekend forced me to defend every architectural choice why Kafka and not a simple queue, why Claude Vision and not cheaper OCR, why caching is so important. Some of those answers didn't hold up, and the ones that didn't became lessons I've carried straight into what I'm building now. For the moment Massiv Cart sits parked as a portfolio piece; the infrastructure is there if I come back to it. The more valuable thing is what it taught me about building for a market everyone else has written off.