AI Math Mentor

Inspiration As a parent, I noticed a recurring challenge at home: helping my kid with math requires a lot of time, and students nowadays tend to rely on standard AI tools just to get quick, ready-made answers. According to Estonian statistics, student performance in graduation math exams has been declining, and schools are facing a serious shortage of math teachers.

This inspired me to build Math AI Mentor – an individual, patient AI mentor designed not to give away answers, but to build student confidence and deep understanding through guided learning.

What it does Math AI Mentor acts as a personalized teacher for students from grades 1 to 12. Instead of solving problems for the student, it uses the Socratic method to ask guiding questions, check steps, and adapt to the student's unique way of thinking, backed by official national curriculum materials.

How I built it (Solo Build) As a solo builder, I designed and deployed the entire system from scratch using a multi-agent hierarchy:

Math AI Mentor (Root Agent): Acts as an expert Socratic math tutor for grades 1–12, guiding students without giving direct answers, handling session flow via instructions and end_session.

Curriculum Agent: Manages curriculum standards and difficulty levels based on the Estonian national math curriculum (grades 1–12) to guide the root agent, utilizing the search_curriculum tool, instructions, and end_session.

Calculation Agent: Performs precise mathematical computations and verifies algebraic steps requested by the root mentor agent, featuring instructions and end_session.

Google Cloud Platform & Vertex AI: Utilizing cloud infrastructure and a secure Data Store to host curriculum data.

Core Safety & Optimization: Configured with low temperature settings to prevent hallucinations, secure IAM permissions, and token optimization.

Multilingual Support To address the diverse educational landscape in Estonia, Math AI Mentor fully supports Estonian, English, and Russian. The system automatically detects the student's language and guides them using the Socratic method in their native or preferred language.

Challenges I faced Managing the entire project lifecycle alone—balancing the multi-agent workflow between the Math AI Mentor, Curriculum, and Calculation Agent to prevent token waste, ensuring strict adherence to the Socratic rule (never giving direct answers), and setting up secure Google Cloud infrastructure entirely on my own.