Keywords
AR, Object Detection, ML, Inference, Grocery

Technologies
Android, YOLOv8, TensorFlow Lite, FastAPI, ML Kit

GrocerARy is an Android application that brings augmented reality and real-time machine learning to the everyday task of grocery shopping, overlaying product information directly onto the live camera view without disrupting the user experience.

At its core, the app supports two inference modes: on-device detection using a custom-trained YOLOv8 model exported to TensorFlow Lite, and server-side inference via a FastAPI backend with GPU acceleration. This dual-mode architecture enables a structured evaluation of the trade-offs between portability and raw performance, with optimisation strategies including frame skipping, resolution scaling, and confidence thresholding applied across both modes.

Beyond object detection, the app integrates barcode scanning through Google ML Kit, real-time nutritional data retrieval from the Open Food Facts API, and a persistent shopping list that visually highlights matching products in the camera view.

The project demonstrates that capable, low-latency AR inference is achievable on consumer Android hardware, while server-side processing offers a meaningful performance ceiling for environments where connectivity is reliable.