InsightUI

Inspiration

We wanted to build something that connects the bridge between web users that might have a harder time than others. With a continually growing online space, it is important we keep these users in mind and to ensure every website is built for every user. Google's Lighthouse Report was a driving factor for inspiring this project.

What it does

InsightUI is a real-time attention-driven UX optimization tool for developers.

Here’s how it works:

Developers assign importance rankings to their website’s HTML components. Using an eye tracker, InsightUI measures where users’ attention actually goes. If users drift away from the most important elements, we calculate an attention score and flag the mismatch. This data is sent to our Gemini-powered LLM, which analyzes the issue and suggests immediate, targeted changes to bring user focus back where it belongs. Developers can even apply these fixes to their site in real time. Once a session is complete, a second Gemini model reviews all the attention data and LLM explanations, then generates a developer-friendly performance summary — highlighting strengths, weaknesses, and the reasoning behind revisions. The result: Developers gain actionable insights into how well their site captures user attention, along with AI-driven improvements they can implement instantly.

How we built it

Backend - Built mainly from Python Frontend - Python Tkinter AI-Integration - Gemini API - 2 models working together to optimize components in real time and give a feedback summary. Database - MongoDB

Challenges we ran into

One of the biggest challenges we faced was working with the eye-tracking data in real time and translating raw gaze points into passable data. It wasn’t enough to know where a user was looking — we also had to reliably map those gaze positions back to specific HTML sections of the webpage. This required fine-tuning both the tracking pipeline and the way we segmented components on the page. We also worked extensively on calibrating the open-source code of the tracker so we would get the most consistent data points. On top of that, we ran into difficulties structuring our prompts for Gemini: we needed responses that weren’t just general advice but concrete, usable HTML revisions. We also had to conquer all these issues while keeping token usage efficient, balancing detail against cost and speed.

Accomplishments that we're proud of

Were extremely proud of reaching end goal of a working demo has a full eye tracking system, 2 LLM's, and updates the page in real time instead of just suggestions the developer can make. Our goal was to streamline this process to encourage developers to continually support accessibility in their pages and gain clarity in how to design for it in mind.