Skip to content
PrakharGoel7Public

About

Live website research: a Tavus AI interviewer observes friction, coding agents build a fix, and the same participant tests it.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Loop

See the friction. Change the site. Test it again.

Live website research with a face-to-face AI interviewer and a coding agent.

Try the demo · Architecture · Run locally

Checks React TypeScript Tavus

Loop's live website research landing page

Overview

Loop is a dynamic user-testing platform that lets people test website changes during the same interview in which they request them.

A Tavus AI interviewer observes the participant's shared screen, listens to feedback, and asks questions about their experience. When the participant suggests an improvement, a coding worker builds the change in the background. The interviewer then invites them to try the updated website and share their reaction.

The demo uses a clothing store to show this workflow.

Architecture

Loop architecture: user, Tavus CVI, knowledge base, coding worker, test website, and website developer

Try it

Hosted: Open Loop. No invite code is required. Use a desktop browser, allow camera and microphone access, and share the store tab or window when asked. Checkout is simulated; no real payment is needed.

Local: Node.js 22.12+, pnpm 10, Git, an authenticated Codex CLI, and a Tavus API key are required for the full experience. FFmpeg prepares friction clips.

git clone https://github.com/PrakharGoel7/loop.git
cd loop
pnpm install --frozen-lockfile
cp .env.example .env
# Set LOOP_CODEX_BIN to your Codex executable in .env.
pnpm dev:all

Open http://127.0.0.1:5173 and configure Tavus. See setup instructions for environment variables and knowledge-base configuration.

Built with

React · TypeScript · Vite · Express · Tavus CVI · Daily · Codex CLI · FFmpeg · Google Cloud Storage

The hosted app and coding worker run on Google Compute Engine.

About

Live website research: a Tavus AI interviewer observes friction, coding agents build a fix, and the same participant tests it.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages