personavision - Swarm of AI agents that test UX before real users

PersonaVision tests your website with a crowd of AI people before real users ever see it. You describe your audience in a sentence, for example "software engineers aged 18–30 building their own products, mostly on laptops". Each AI persona then uses your live site in a real browser, on a phone or a desktop, to reach a goal you set in plain words. Every session leaves a trajectory: a video recording, where the person looked and in what order, what they clicked, where they gave up, and the reason behind every move, in their own words. Across the audience these add up to attention heatmaps, common paths, a funnel showing where people dropped out, and time on each step. Then Muse Wonder reads all the sessions and makes UI change and retests on same audience to make better UX The repo also includes a webcam eye tracker that runs locally. Recordings of real people become few-shot gaze examples for the agents or data for training small models of real behaviour.