# MathMate AI - **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon) - **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST) - **Where:** Online - **Team:** [Anojaa Gnaneswaran](https://cerebralvalley.ai/u/Anojaa) - **GitHub:** https://github.com/anojaagnanes/mathmate-ai.git - **Demo video:** https://youtu.be/IMvKsh4gJ7w?si=cjIf_wcgp9Xvdbq9 - **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery - **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/524 MATHMATE AI — DON'T JUST SOLVE IT. UNDERSTAND IT. One Meta AI Model. Seven Specialised Capabilities. Multiple AI Agents. One Personalised Learning Journey. MathMate AI is an agentic, bilingual visual mathematics tutor that transforms students' mistakes into opportunities to understand, practise, and build confidence. THE PROBLEM Many students struggle with mathematics because traditional learning tools provide correct answers without identifying where their reasoning went wrong. Repeated mistakes can reduce confidence, especially when students hesitate to ask questions or need explanations in their native language. OUR SOLUTION MathMate AI helps students understand not only what went wrong, but why. Students can upload handwritten solutions, receive mathematically verified feedback, and learn with Mia, an interactive 3D AI tutor who provides supportive explanations in Tamil and English. The system connects mistake detection, personalised practice, visual learning, and gamified challenges into one learning experience. KEY FEATURES • Check My Work: Interprets handwritten mathematical solutions, allows students to confirm extracted steps, and uses SymPy to identify the first detected mathematical mistake. • Autonomous Mastery Mission: Analyses verified mistake history, identifies weaker concepts, and coordinates personalised learning activities through an agentic workflow. • Mia Learning Studio: Offers interactive visual explanations, hands-on mathematical exploration, a 3D teacher avatar, and bilingual conversational learning. • Math Treasure Journey: A five-stage gamified adventure with mathematical challenges, progressive difficulty, hints, collectible gems, and targeted practice. HOW WE USE META AI — ONE MODEL, SEVEN CAPABILITIES MathMate AI integrates Meta Muse Spark Open through Baseten. Instead of deploying seven separate AI models, we reuse one Meta model across seven specialised capabilities: 1. Handwriting Transcription: Interprets mathematical working from uploaded images. 2. Question Interpretation: Understands students' mathematical questions and learning requests. 3. Bilingual Teaching: Generates step-by-step explanations in Tamil and English. 4. Real-World Examples: Connects abstract mathematical concepts to familiar everyday situations. 5. Practice Generation: Drafts exercises based on students' learning needs. 6. Conversational Tutoring: Supports Mia's responses to questions and requests for clarification. 7. Game Narratives: Generates contextual content for Math Treasure Journey. This shared-model approach supports multiple educational functions while keeping the learning workflows coordinated through specialised agents. MULTI-AGENT ARCHITECTURE MathMate AI uses a multi-agent architecture in which specialised agents collaborate across handwriting understanding, mathematical interpretation, teaching, practice generation, and adaptive learning. An orchestration layer coordinates these capabilities according to the student's activity and learning needs. Example workflow: Handwritten Solution → AI Interpretation → Student Confirmation → SymPy Verification → Bilingual Explanation → Personalised Practice Rather than relying exclusively on AI-generated answers, MathMate AI integrates a deterministic SymPy verification engine to check supported mathematical steps and challenge answers. This separates mathematical correctness from AI-generated explanations and helps make feedback more dependable. TECHNOLOGY STACK • AI Model: Meta Muse Spark Open • Inference Platform: Baseten • Architecture: Multi-agent orchestration and specialised learning workflows • Backend: Python, FastAPI, SQLAlchemy • Frontend: Next.js, React, TypeScript • Mathematical Verification: Python SymPy • Interactive Learning: 3D Mia tutor, visual mathematics diagrams, and gamified challenges WHO WILL BENEFIT? • School Students: Receive personalised guidance, visual explanations, and targeted mathematics practice. • Students Hesitant to Ask Questions: Learn in a supportive environment where mistakes become learning opportunities. • Tamil and English Learners: Access explanations in their preferred language. • Students With Limited Tutoring Access: Gain an additional resource for independent mathematics learning. • Parents: Support their children's learning through interactive activities. • Teachers and Tuition Centres: Potentially benefit from future progress tracking, learning insights, and teacher-reviewed exercises. FUTURE GOALS Expand the mathematics curriculum, introduce teacher-reviewed exercises, learning analytics, student progress dashboards, and tuition-teacher support. Our long-term vision is to make personalised bilingual mathematics education more accessible and affordable while complementing human teachers. OUR MISSION Help every student move from “I got it wrong” to “Now I understand.” MathMate AI — Don't Just Solve It. Understand It. ## More from Meta Global AI Developer Hackathon - [Rajat(Self)](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/521) - [Pipeline Copilot](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/522) - [PixyPup](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/523) - [Achmad Royan Qodri Ramadhani](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/525) - [HackX](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/526) - [Rui Codex](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/527) --- Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/524. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt