Ubers File Analyzer

Uber's File Semantic Analyzer gives security analysts a description of what an outgoing file contains before they decide how to handle it. The system extracts text, including text from images, and processes large documents in smaller sections. A language model supplies summaries, relevant entities, and a proposed sensitivity classification.

A separate rules engine uses those outputs to apply policy, while human analysts review significant findings and provide corrections. In Uber's design, the useful change is the evidence available for review: analysts can inspect why a document was flagged and compare the explanation with its contents.