Abinaya VelusamyProgram & Product Manager
CASE STUDY — RESI

Retail Shelf Intelligence.

A computer-vision platform that tells a store manager, in seconds, whether the shelves actually match the planogram — and how long it took staff to put them right.

Best Innovative AI Project GECO Asia Hackathon · 2025
// THE_PROBLEM

A planogram nobody can verify.

Every store runs to a planogram — the plan for exactly which SKU sits where on which shelf. Compliance drives sales, but verifying it is manual and largely guesswork. The store manager we built for couldn't answer four basic questions:

01

Is the store actually adhering to the planogram right now?

02

Are the store keepers restocking the SKUs they're supposed to?

03

How many products are sitting in the wrong place?

04

How long is each staff member spending maintaining the shelves?

// THE_SOLUTION

Point a phone at the shelf.

Staff photograph the shelf in a mobile app. The image goes to a computer-vision service that detects every product on it, compares that against the planogram held in the backend, and returns exactly what needs fixing. Managers see the aggregate on a dashboard.

SHELF_SCAN.LIVE
01

Capture

Staff photograph the shelf in the mobile app — no special hardware.

02

Detect

A YOLO model plus a vision-language model identify every product and its position.

03

Compare

Detections are checked against the store's stored planogram.

04

Act

The app returns a worklist: what to restock, what's misplaced.

05

Measure

Compliance and staff time roll up to a manager dashboard on web or mobile.

// ARCHITECTURE

Four services.

MOBILEStaff capture app — React Native, Expo
WEBManager dashboard — React, Vite, Recharts
CV SERVICEProduct detection — Python, FastAPI, YOLO, OpenCV, vision-language model
API SERVICEOrchestration, auth, planogram data — FastAPI, JWT
INFRADocker Compose across all services
// STACK
Python FastAPI YOLO OpenCV Vision-language model React Native Expo React Vite Recharts Docker