MyTutor.ai

MyTutor.ai creates a fun tutor based on what your child loves, using GPT-4.2 and Qloo Insights API to understand their favorite shows, movies, and games to create personalized math challenges.

Comment

Inspiration

This project was inspired by my experience tutoring kids online. As a tutor, I learned how to incorporate my students interests into our math work. Soon I began wondering: could AI be used to truly engage children by understanding what they already love?

What it does

MyTutor.ai uses a large language model (GPT-4.1) alongside the Qloo API to generate a fun, customized avatar and persona for each student’s personal tutor. This AI-powered tutor then creates math challenges tailored to the child’s interests—like pirates, space, ninjas, or animals—and interacts with them in a way that feels playful and relevant. It can even write to a whiteboard, making it much easier for visual learners to understand, and further bridges the gap between what a person and AI agent can bring to the table for young learners.

How I built it

I built it using React, Flask, GPT 4.1, and the Qloo API. To build the avatars, I used the Qloo API to gain insights on what the student likes based on just a few questions and fed that information into an LLM, which (having been given a preset list of possible clothing items) assigned a hat, pair of glasses, skin color, and item to hold in the avatars hand. This same agent also determined the tutor agents personality. This personality (along with instructions for the whiteboard schema) was sent to the tutor agent and informed the way it interacted with the youth. These whiteboard instructions were parsed into a json format the frontend could understand.

Challenges I ran into

It was particularly challenging to get the Ai Agent to consistently write to the whiteboard in a format the frontend could understand. This required constantly improving the schema given to the ai and tweaking the algorithm used to parse the output to be more flexible given how variable LLMs can be when trying to adhere to a specific schema.

Accomplishments that I'm proud of

Making a way for the AI to write to a white board was uniquely challenging but extremely fulfilling. Its seamless now, and really contributes to the feeling that you're talking to and learning from a real person who can change and respond to what you're having difficulty with.

What I learned

I learned a lot about how to have multiple LLM agents interact, and how to facilitate communication between an LLM and other APIs.

What's next for MyTutor.ai

Next, I plan to further develop this tool so that teachers can aggregate classroom-wide preference and learning data, helping them adapt lessons to students’ collective interests and learning challenges. I also want to support teacher-defined study plans that the AI tutors can use to guide individual students toward shared class goals. I believe this approach could transform large classrooms by giving each child a tailored experience—ensuring no one falls behind simply because they weren’t engaged.

Built With

Share this project:

Updates