groundnote demo
A doctor–patient transcript goes in. A 239M model trained from scratch writes the note, and a lexical checker underlines every clause by whether its words appear in the transcript.
Everything runs in your browser. Your text never leaves this page. The first run downloads the model (about 480 MB, cached after that) and a Python runtime for the checker. A long visit takes about a minute. Research prototype, not for clinical use: do not paste real patient information.