About the Project
Inspiration
UXStream Auditor was inspired by a recurring problem in product teams: UX issues are often detected late, subjectively, or only after users complain. Watching session recordings is useful but time-consuming, repetitive, and hard to scale. The idea was to let AI do the boring part: observe, analyze, and point out potential UX friction automatically.
How We Built It
The project is a web application where users upload a video of a user interacting with an interface.
The frontend is built with React + Vite, providing a simple flow for video upload and report visualization.
The backend is developed in Python with FastAPI, exposing a REST API that processes the video and sends structured prompts to the Gemini API, which performs the UX analysis and returns actionable insights.
The system focuses on identifying patterns such as hesitation, repeated actions, confusion points, and flow breaks, then translating them into human-readable UX recommendations.
Conceptually, the goal can be summarized as minimizing UX friction:
[ UX_{friction} = \sum_{i=1}^{n} (hesitation_i + repetition_i + error_i) ]
What We Learned
- Designing good AI prompts is as important as writing good code.
- Clear UX insights require structured, constrained outputs from the model.
- Handling large media files in a web app introduces performance and CORS challenges that must be addressed early.
- Simplicity in UI greatly improves trust in AI-generated results.
Challenges Faced
- Managing video uploads and processing without degrading user experience.
- Dealing with cross-origin (CORS) issues between frontend and backend deployments.
- Ensuring the AI feedback was useful, specific, and not generic UX advice.
- Balancing speed, accuracy, and cost when calling AI APIs.
Despite these challenges, the project proved that AI-assisted UX analysis is not only feasible, but highly valuable when focused on clarity and actionability.
Log in or sign up for Devpost to join the conversation.