Embodied vocal assistant
A vocal assistant improving access to online resources for impaired users — accessibility-first by design.
This project began from a plain observation: a large amount of what makes the web usable — dense navigation, hover states, visual hierarchy — actively works against users with motor or visual impairments. The embodied vocal assistant is an accessibility-first attempt to make everyday online tasks reachable by voice, with the interface built around the conversation rather than bolted on afterward.
It pairs speech recognition with an intent layer that maps spoken requests onto concrete page actions, and it speaks back with enough context that a user never has to see the screen to know what happened. Designing it accessibility-first — rather than retrofitting a voice layer onto a visual app — meant the whole flow, from error recovery to confirmations, is legible without sight.
In task-completion testing the assistant cut the time to finish common flows by roughly 40% for the users it was built for — the clearest signal that voice-native, not voice-added, was the right starting point.