Bower

Solution

Bower uses Google Cloud to address this challenge: Bower needed to move beyond barcode-dependent recycling guidance so users anywhere could identify waste items, understand local disposal options, and keep the app learning as it expanded globally. Vertex AI hosts and serves the machine learning models Bower uses to identify waste items from photos, serving as the serving layer. Compute Engine provides Cloud GPU capacity for training the image-identification model at global scale, serving as the compute layer. Vision AI performs object detection to identify item type, material, estimated mass, and brand from user-submitted images, serving as the compute layer. Gemini processes unstructured user input to evaluate recycling drop-off points and continuously enriches the global waste knowledge base, serving as the compute layer.

Data flow

Users upload a photo, Vision AI and the Vertex AI model identify the waste item and its characteristics, Compute Engine GPU training keeps the recognition model current, and Gemini processes additional user input about recycling locations so the knowledge base expands market by market.