ShelfSense AI

ShelfSense AI transforms CCTV into real-time inventory intelligence. We track shelf-side behavior to predict demand and slash waste, capturing insights that traditional registers miss.

Comment

Inspiration

The retail world is currently flying blind. Most inventory systems only know a product is gone once it hits the scanner at checkout, but they have no idea what happens on the shelf itself. We wanted to bridge that gap by giving store owners a live pulse on customer behavior. The goal was to build something that stops waste before it starts and understands why a product is being picked up but put back down.

What it does

ShelfSense AI is an intelligent command center for retail owners. It uses AI powered CCTV analytics to track more than just sales. The system monitors: Pick versus Return counts to see if customers are interested in a product but rejecting it after reading the label. Real time stock levels that automatically flag items as Low Stock or Overstock based on optimal targets. Expiry tracking that warns when perishable items like Organic Whole Milk are nearing their end. Visual Heatmaps and Live Feeds to show exactly where the action is happening on the floor. AI Insights that use the Gemini API to analyze this data and give the owner actionable advice.

How we built it

We leaned into a modern tech stack to keep the interface snappy and the data reactive. The frontend is built with React 19 and Vite for a lightning-fast development experience. We used Tailwind CSS to create a clean, professional glass-panel aesthetic that feels like a premium enterprise suite. For the brains of the operation, we integrated the @google/genai SDK to process product interactions and generate demand predictions. Recharts handles the heavy lifting for data visualization, turning raw numbers into readable trends. The logic uses React hooks like useCallback to manage complex state updates across the dashboard.

Challenges we ran into

One of the trickiest parts was managing the state of an entire store in a way that felt "live" without constant flickering. Ensuring that a restock action or an expiry update propagated correctly through the Dashboard, Inventory List, and Heatmap views required careful architectural planning. We also had to ensure the Gemini API could handle the specific context of our product data to provide insights that were actually useful rather than generic.

Accomplishments that we're proud of

We are particularly proud of the AI Insights engine. It is one thing to show a graph, but it is another to have an AI tell you that your avocados are overstocked and suggest a flash sale. The Live Operations header and the seamless transition between the high-level overview and granular inventory lists create a workflow that feels very natural for a busy store manager.

What we learned

This project taught us that visual data is a massive untapped resource in retail. Traditional Point of Sale (POS) data is a lagging indicator, while customer-product interactions are leading indicators. We also got a deep dive into React 19 and the power of the Gemini API in transforming raw inventory metrics into a conversational, helpful assistant.

What's next for Shelfsense AI

We want to move beyond just tracking and into active automation. The next step is integrating direct camera access through the browser to allow owners to test the CCTV logic in real time. We also plan to expand the Owner Suite with automated reordering systems that talk directly to suppliers when the AI predicts a stockout.

Built With

Share this project:

Updates