Neuro-Math
An AI-powered math solver that reads photos of handwritten problems and returns detailed step-by-step solutions using Google Gemini Vision API.
About This Project
Neuro-Math is a personal project that uses Google Gemini's Vision API to interpret photos of handwritten math problems and return step-by-step solutions. Users can upload or capture a photo of any handwritten equation or problem, and the app sends it to the Gemini Vision model which reads the handwriting and generates a structured, step-by-step breakdown of the solution. The frontend is built in React and the backend is a Node.js API that handles image processing and Gemini API communication.
Key Features
- Handwritten math problem recognition via Google Gemini Vision API
- Step-by-step solution generation from raw image input
- Photo upload and camera capture support on the React frontend
- Node.js backend handling image relay and Gemini API integration
- Supports algebra, arithmetic, and basic calculus problems
Challenges & Solutions
Prompt engineering Gemini to return consistently structured step-by-step outputs
Crafted a strict system prompt instructing Gemini to respond only in a numbered JSON array of steps, then validated and parsed the response on the Node.js backend before sending it to the frontend — falling back to a plain-text display if the structure was missing
Handling varied handwriting quality and image resolution gracefully
Added server-side image pre-processing (resize to a fixed max resolution, convert to JPEG) using the Sharp library before forwarding to Gemini, reducing API errors on low-quality camera captures