Receipt and Invoice Text Reading

Cloud Vision reads printed and handwritten text out of an image using Google's pre-trained models. A Beam agent scans an uploaded receipt or invoice image, reads the vendor name, date, and total, and posts those fields into the expense or accounting system where a matching record does not already exist. It checks the read amount against the file before posting. Images that are blurry, cut off, or where the model returns a low confidence score are set aside, and someone on the finance team opens the original image to enter the details by hand.

Object and Label Detection

Cloud Vision can tell what objects, scenes, or labels appear in a photo. A Beam agent reads a newly uploaded photo, such as a product image or a site inspection picture, checks it against a list of labels a team has approved, such as damage, missing safety gear, or a specific product category, and tags the record accordingly in the system it came from. It only applies labels already on the approved list. Photos where no label meets a confidence threshold, or where more than one conflicting label applies, are sent to a person to tag by hand.

Unsafe Content Detection

Cloud Vision can check an image against Google's safe search categories for adult, violent, or otherwise unsafe content. A Beam agent reads an image at the point it is uploaded to a public area, such as a community page or product listing, and checks it against the safe search result. Images that come back clearly within a category a company's policy already treats as unsafe are held out of the public listing before it goes live. Anything the model marks as possible but not certain, or a category the policy does not clearly cover, is sent to a moderator to decide.