17MB model beats human experts at pronunciation scoring
Beats humans at pronunciation scoring but doesn't ship product integration yet.

Personal email sign ups have a limited demo, and logging in with one of 80 schools will allow you full access, including able to post to the fora yourself.
I would greatly appreciate any (and literally ANY) feedback you have!! :P
Beats humans at pronunciation scoring but doesn't ship product integration yet.
StackExchange meets AI math tutor, but MIT-only access kills momentum.
Beats human experts at phoneme scoring while 70x smaller than SOTA models.
Structural OCR for handwritten math—preserves alignment, tables, equations where standard OCR fails.
Unnamed drafts and automatic resurfacing are the two gutsy choices here — start writing and the app tries to pull up related notes for you. It's a neat UX shortcut if the semantic matching is strong, but the landing page hides how recall is implemented (model, privacy, offline), and in a crowded space with Mem/Reflect clones that detail matters.
Markdown-to-mind-map rendering is slick, but Obsidian Canvas already maps local notes better.