Three-layer machine-learning system drafts sentences for non-verbal users; ConnectAble tops the field
By NIPUN SAINI · The Composing Room
ConnectAble, a predictive augmentative and alternative communication system, took first place honors this season. The build stacks a three-layer machine-learning cascade that drafts full sentences for non-verbal users, cutting the taps required to speak from dozens to a handful.
The system predicts what its user wants to say next and offers whole utterances, not letters — a bet that context beats keystrokes. Judges awarded it the top prize. The full case study, including the cascade architecture, appears on its own page.
Linx runs its intelligence entirely offline, keeping watch in Alzheimer's care settings without shipping a byte to the cloud. The build placed fourth in national competition.
A healthcare model reading outcomes across 75 demographic subgroups reports 93.7 percent accuracy, earning an honorable mention and surfacing disparities hidden in aggregate statistics.
Sea Score judges shoreline cleanups with zero-shot computer vision — no task-specific training required — and collected an honorable mention for the trouble.