Vision Homework Tutor

Inspiration

I am an engineering student, and students are probably one of the groups that use AI the most today. While AI tools help a lot, I noticed that many students have started depending on them too much. Instead of understanding the concept, they often just copy and paste answers from AI for homework or assignments. I wanted to build something that helps students actually learn the concept instead of just getting the answer. While using tools like ChatGPT, I also noticed some small problems. For example, when uploading an image, there was no option to crop only the part of the question I wanted to ask. Also, it did not have a simple way to download the chat for revision later. Students also had to switch to other platforms to take tests or practice questions. These small problems gave me the idea to build Vision Homework Tutor, a tool that not only solves problems but also helps students understand them better.

What it does

Vision Homework Tutor is a multimodal AI learning assistant that helps students solve and understand homework. Students can ask questions using: text,images,camera,cropped image sections,PDFs

The system analyzes the question using Gemini AI and explains the solution step by step. Some important features include: ----step-by-step tutor mode ----image cropping before sending to AI ----multilingual explanations ---dual language explanation option ---downloading chat history for revision ---practice question generation ---mistake analysis from student attempts ---full screen MCQ exam mode ---instant exam analysis with explanations The goal is to make AI act more like a learning tutor instead of just an answer generator.

How we built it

I built the project using a full-stack setup. For the frontend, I used: ---React + Vite ---Tailwind CSS for styling ---Framer Motion for animations For the backend, I used: ---Python ---FastAPI The main AI processing is done using the Gemini API, which handles image understanding, question solving, and explanation generation. I also used OCR to extract text from images when needed.

Challenges we ran into

One challenge was handling homework images that contained many questions. Sometimes the AI would analyze the entire page instead of the specific question. To solve this, I added an image cropping feature so users can select exactly the part they want to send. Another challenge was making explanations feel more like a tutor instead of a simple AI answer. I had to experiment with prompts and response formats to generate clearer step-by-step explanations. Handling OCR errors from images was also difficult, so I added preprocessing and fallback methods before sending the content to Gemini. Designing the exam mode with timer, full-screen mode, and tab-switch detection also required careful frontend logic.

Accomplishments that we're proud of

I am proud that the project became more than just a homework solver. Some things I am proud of include: ---building a multimodal homework tutor ---implementing image cropping for specific questions ---creating a step-by-step tutor mode ---adding a full exam simulator with analysis ---supporting multiple languages and dual-language explanations ---allowing users to download chat history for revision The project now feels like a complete AI learning assistant instead of a simple chatbot.

What we learned

Building this project helped me learn a lot about working with multimodal AI systems. I learned how to combine image processing, OCR, and AI reasoning into one pipeline. I also gained experience using the Gemini API for solving real problems. Another important lesson was that good user experience matters a lot. Small features like cropping images, step-by-step explanations, and practice tools can make AI much more useful for students.

What's next for Vision Homework Tutor

There are many improvements I would like to add in the future. Some ideas include: ---personalized accounts for each to store and track their progress and all ---better handwriting recognition ---voice conversation with the tutor ---personalized learning paths ---tracking student progress over time ---more advanced exam simulations The long-term goal is to make Vision Homework Tutor a complete AI study companion that helps students learn concepts deeply, not just get answers.

Built With

• docker
• fastapi
• framer-motion
• google-cloud-run
• google-gemini-api
• html
• javascript
• ocr
• python
• react
• react-easy-crop
• tailwind-css
• vite
• web-speech-api