Amazon's Project PI uses generative AI to scan packages and detect defective items

Amazon.com Inc. today revealed one of the more practical ways it's leveraging generative artificial intelligence technology: The company said Project PI is an AI-powered system that combines generative AI and computer vision to reduce the number of defective products leaving fulfillment centers.

Project PI scans each product that passes through an imaging tunnel, using advanced algorithms to detect issues such as damaged packaging, the wrong color or size, or expired goods.

If a defect is found, the item is blocked by the system and not shipped to the customer. Amazon's team inspects the flagged items to see if they are truly defective and decides what to do with them – whether to discount them for resale on the Second Chance platform, donate them, or find another use for them.

Project PI, which means “private investigator,” is already up and running in some of Amazon's fulfillment centers, where the company says it scans millions of products that pass through its imaging tunnels every day. As each product goes through the tunnel, the system scans it for defects in a variety of ways.

For example, optical character recognition models check product expiration dates to ensure that expired items are not sent to customers, computer vision models trained on reference images from Amazon's product catalog and other products sent to customers analyze color to ensure it matches the customer's requirements, and any visible physical damage to the product, like a bent book cover or ripped packaging, is also detected.

Amazon says catching defective products before they are shipped improves customer experience while also reducing the number of product returns that increase carbon emissions and packaging waste, strengthening its sustainability efforts.

Dharmesh Mehta, Amazon's vice president of Worldwide Selling Partner Services, said it's important to deliver the right experience every time customers shop, so the technology is a win-win for Amazon customers, its selling partners and the environment.

The Project PI system can also help Amazon investigate the root causes of poor customer experiences. When an item is returned, Amazon uses the multimodal LLM system to review customer feedback and analyze images taken by Project PI and its fulfillment centers to see what went wrong. Project PI can then learn from the mistake, and Amazon's selling partners can also access the data on the defective item to prevent similar mistakes in the future.

Amazon said it is committed to continuously improving its operations and sustainability efforts, and plans to expand Project PI to its other fulfillment centers later this year.

The effort is just the latest AI tool Amazon is using to become a more sustainable company: For several years, the company has been using AI models to scan products to reduce their use of packaging, monitor product quality in its grocery service and estimate the carbon footprint of each product.

Additionally, the company's cloud computing services division, Amazon Web Services, has emerged as a key platform for AI developers, offering access to models from companies such as OpenAI as well as tools for building AI-powered applications.

Image: Amazon.com