A tool that uses AI to generate and review alt text at scale, so large PDF and Word documents can meet ADA accessibility requirements without manually captioning every image by hand.
Documents with hundreds of embedded images are common in academic and government publishing, and captioning each image by hand for ADA compliance does not scale. This project builds a pipeline that extracts every image from a PDF or Word document, generates a caption or alt text description with an AI model, and lets a reviewer approve or edit each one before it gets written back into the original document.
I am building the full pipeline: extracting images from PDF and Word source files, sending each image to an AI API for caption generation, and building a review interface where a human can approve or modify a caption before it is saved. The core design constraint is that a human stays in the loop before any alt text ships, since an unreviewed AI caption that is wrong can be worse than no caption at all for a screen reader user.