Frictionless

Landing page
Personas going through site
View / Select Your Persona
View Report Findings

Inspiration

Teams often wait for enough traffic before they can learn why users leave. We wanted to make early UX feedback faster by showing how different customer personas experience a site before launch.

What it does

Frictionless runs a cohort of AI personas through any website. It tracks their navigation, clicks, dwell time, mouse movement, and friction points, then turns that evidence into concise UI recommendations and a safe live preview.

How we built it

We built the dashboard with Next.js, React, TypeScript, and Tailwind. Browserbase and Stagehand run isolated cloud-browser sessions in parallel. Moorcheh provides persona context, OpenAI reviews browser evidence for design issues, and SQLite stores runs locally.

Challenges we ran into

Keeping live browser sessions visible long enough, handling sites that block automated browsing, and making sure recommendations are based on real evidence instead of generic AI feedback. We also had to balance realistic agent behavior with safe interactions that do not submit forms or make purchases.

Accomplishments that we're proud of

Without modifying the original website. The product can clearly show both what personas did and why a change is recommended.

What we learned

Browser automation alone is not enough; the useful part is connecting interaction evidence to a clear, specific recommendation. Context management, browser-session reliability, and concise presentation matter as much as the AI model.

What's next for Personalize

We want to add durable cloud storage, more persona sources, richer telemetry such as scroll heatmaps and eye-tracking proxies, and side-by- side testing of multiple combined UI fixes.