Math Tutor

Opportunity

Many students, especially in high school and early university, struggle to ask questions in math class. Fear of embarrassment, fast-paced lectures, or anxiety about “not getting it” leaves gaps in learning that grow over time. Traditional tutoring is expensive and not always available when a student gets stuck. The opportunity was to create a self-paced, judgment-free environment where any learner could ask any question, no matter how simple or advanced, and receive clear, visual, step-by-step help.

I wanted to design a system that reduces this emotional barrier using approachable humanlike interaction, while maintaining high-quality mathematical reasoning and visual explanation.

Solution

I built an AI math tutor that can not only answer questions but also explain concepts visually and conversationally, to simulate an understanding and patient tutor. The technologies used are listed belowed:

Gemini Live (Voice Agent):

  • Powers real-time, two-way spoken conversations.
  • Enables students to ask questions naturally and receive instant spoken responses.
  • Chosen for its advanced AI conversation abilities and multimodal input support, essential for interactive math tutoring.  

Pipecat (Voice Fine-Tuning and Orchestration):

  • Refines the AI’s speaking style, tone, pacing, and empathy, making interactions feel humanlike.
  • Manages real-time voice and video streaming.
  • Provides precise control over conversation flow and context to keep tutoring sessions smooth and natural.

Gemini 2.5 Flash (Diagram Generation):

  • Creates dynamic diagrams like graphs, proofs, and geometric shapes on the fly.
  • Adds a visual learning component that helps students understand complex math concepts better.

Python Backend (AI Pipeline and Server):

  • Runs the core tutoring system, managing audio streams and coordinating AI model calls.
  • Integrates the AI services, handles conversation context, and processes multimodal inputs (voice, images).
  • Chosen for its strong AI ecosystem, ease of prototyping, and suitability for complex real-time workflows.

TypeScript Backend (Logic and Integration Layer):

  • Implements the frontend logic that interacts with AI models, structures prompts, and manages dialogue context.
  • Provides strong typing and tooling to avoid bugs and improve maintainability.
  • Handles frontend/backend communication and real-time user interactions.

‍

Together, these technologies create an AI math tutor that listens, understands, explains visually, and adapts to each student's learning needs in real time. Python powers the backend AI orchestration, TypeScript ensures a robust, maintainable frontend and integration, and Gemini Live and Pipecat deliver the intelligent conversational and audio experience. Gemini 2.5 Flash adds the vital visual explanation layer.

Impact

The AI math tutor bridges the confidence gap between students and learning. It creates a safe space where curiosity is encouraged and mistakes are normalized.

  • For students: freely ask any question without judgment, get immediate corrective feedback and visual support.
  • For educators: a scalable supplement that frees up time for deeper mentorship.
  • For me as a developer: deepened my understanding of conversational AI latency, multimodal model orchestration, and empathetic interface design.

More Information

Learn More