Voice driven tools make documents more accessible
Assistivox is an accessible document reader that helps users navigate longer documents and find relevant passages through semantic search. It combines a reader interface with speech playback, chunk-by-chunk navigation, and search results that can be highlighted in the source text.
I developed Assistivox for research at the University of Georgia under Dr. Bill Hollingsworth. My goal was to make document content easier to access, especially for people who benefit from spoken content or need to locate a particular idea without scanning every page. By pairing semantic search with reader navigation, users can move from a relevant result to its place in the document.
One of the main challenges was turning different document formats into useful search chunks without losing where each passage came from. The backend extracts and indexes document text, while character offsets and page information connect each search result back to the reader for highlighting and navigation.
This project helped me understand how semantic search depends on the entire pipeline, from text extraction and chunking to embeddings, indexing, and the reader interface. I also gained experience designing an API and accessible reading interactions that put search results in context instead of presenting them as isolated matches.