Computer Vision
ViaTouch's VICKI machines depend on a computer-vision system that can understand what is stocked, what moved, and what a shopper actually selected. rustLabs supported that training loop by turning field footage and machine outputs into cleaner examples the model could learn from.
The work centered on product recognition and shelf-state analysis: identifying the right item when packaging was similar, lighting was imperfect, a product was partly blocked, or the shelf no longer matched the expected layout. Those edge cases matter because the machine has to make the right call in real retail conditions, not just in clean test footage.
Better vision data, better checkout decisions.
By reviewing detections, correcting uncertain labels, and clarifying when the system should treat an item as unknown, rustLabs helped create training signal for more reliable item analysis. The goal was simple: help VICKI distinguish stocked products, recognize movement on the shelf, and reduce the kinds of mistakes that can create bad checkout events or inventory drift.
Guardrails for a smarter retail assistant.
VICKI also includes an AI assistant that can answer questions about the products inside the machine. rustLabs helped shape the guardrail logic and reference-product data behind that assistant so responses stay grounded in what is actually available. That improves security, keeps answers on topic, and gives the model a safer way to decline when a shopper asks for something outside the stocked catalog.
Together, the vision review and assistant guardrails make the system more dependable: better product understanding, cleaner model feedback, stronger inventory confidence, and more useful conversations at the machine.