Inspiration
We were inspired by a simple problem: students often know they need help, but getting personalized tutoring can be expensive or unavailable. Most educational AI tools focus on the final answer instead of understanding how a student arrived there. We wanted to build something different—an AI tutor that watches the student's work and helps them understand their mistakes without immediately giving away the solution.
What it does
Lumina is an AI-powered tutoring platform designed around how students actually solve problems. Students can write their work on a digital board and ask Lumina for help when they get stuck. Instead of just providing the correct answer, Lumina analyzes the student's work, identifies the misconception behind a mistake, and gives a guiding question or hint to help the student figure it out themselves.
Lumina also supports tutoring sessions and is designed to work across mobile and web experiences, making personalized learning more accessible.
How we built it
We built Lumina with a mobile app, web frontend, and backend working together. The frontend provides a pen-first workspace where students can write and interact with lessons. The backend uses FastAPI to process the student's screen image and geometry information and return tutoring hints and visual targets.
For the AI layer, Lumina separates vision, visual pointing, and reasoning into different services. This allows the system to understand what the student wrote, identify the relevant part of their work, and generate a Socratic hint instead of simply revealing the answer. We also added session management and security features such as authentication, rate limiting, image-size protections, and controlled API access.
Challenges we ran into
One of our biggest challenges was getting the AI to understand a student's process rather than just their final answer. Handwritten work can be messy, and the system needs to determine what the student is doing before it can give useful feedback.
We also had to make the different parts of Lumina communicate reliably—the handwriting interface, AI vision, reasoning system, and backend all need to work together in real time. Making the experience feel natural while also keeping student data and sessions secure was another major challenge.
Accomplishments that we're proud of
We're proud that we built more than just an AI chatbot. Lumina is designed around the student's actual work. A student can write out their reasoning, have Lumina identify where their thinking went wrong, and receive a hint that helps them continue solving the problem themselves.
We're also proud of building the project as a full system with mobile, web, and backend components rather than relying on a single interface. The backend includes protections for authentication, rate limiting, image uploads, and session privacy.
What we learned
We learned that effective AI tutoring is not just about having a powerful model. The way information is presented matters just as much. Giving students the answer may solve one problem, but helping them understand why they made a mistake can help them solve future problems independently.
We also learned a lot about connecting AI models with real user interfaces, processing visual information, handling handwritten input, and designing APIs that are secure enough for real applications.
Log in or sign up for Devpost to join the conversation.