Euler - Unblind the Math Genius in you

Capture Image
Home page
Explanation 1
Explanation 2

Inspiration

Numerous students struggle with math not because they lack intelligence, but because they lack accessible and interactive guidance when solving problems. We wanted to build a system that feels less like a calculator and more like a real tutor that can see, explain, and discuss math naturally.

We were also inspired by the idea of making mathematics more accessible for everyone, especially students who feel intimidated by equations or need verbal explanations to better understand concepts. Instead of static answer generators, we wanted conversational learning with reasoning, textbook grounding, and voice interaction.

That inspiration led to Euler — “Unblind the Genius Math in You.”

What it does

Euler is an AI-powered math tutoring platform where users can upload a picture of a math problem and immediately receive:

A transcription of the equation/problem using AI OCR
A step-by-step explanation
A conversational tutor experience through text or voice

The platform supports:

Real-time image-to-math understanding
Follow-up conversations for each problem
Voice interaction with speech-to-text and AI-generated spoken responses
Retrieval-augmented generation (RAG) using a real OpenStax textbook
Phone camera capture through a QR-code workflow
Persistent problem history and retrieval traces

Euler can intelligently decide when to search the textbook for deeper conceptual explanations and when simple reasoning is enough.

How we built it

We built Euler using:

Next.js 16 App Router
TypeScript
Tailwind CSS v4
shadcn/ui
MongoDB Atlas Vector Search
Google Gemini 2.5 Flash
ElevenLabs TTS

AI Pipeline

Users upload an image of a math problem
Gemini performs OCR transcription
A second Gemini pass generates the first explanation
A conversational AI agent continues the tutoring session
The agent can call a custom searchTextbook tool that performs Atlas $vectorSearch retrieval over OpenStax textbook embeddings

Voice System

We implemented:

Browser-based speech recognition
Automatic silence detection and auto-submit
Natural speech rewriting for LaTeX expressions
ElevenLabs voice synthesis for spoken responses

Database + RAG

MongoDB Atlas stores:

Problem conversations
Uploaded images as Binary documents
Vector embeddings of textbook chunks

The retrieval system uses:

Gemini embeddings
Atlas Vector Search
Chunked OpenStax Algebra & Trigonometry textbook data

Challenges we ran into

We worked hard to identify the fastest and simplest way to capture math problems in real time while still maintaining high OCR accuracy and conversational quality.

Some major challenges included:

Making OCR reliable across handwritten and printed equations
Preventing hallucinations during math explanations
Designing a clean agent loop that knows when to retrieve textbook information
Converting LaTeX-heavy responses into natural spoken English
Handling Gemini API rate limits during embedding ingestion
Building real-time voice interaction with silence detection and live audio monitoring
Maintaining strict isolation between problem conversations to avoid context leakage

Another major challenge was optimizing the retrieval pipeline so the AI could quickly search textbook content without slowing down the conversation experience.

Accomplishments that we're proud of

Successfully reading math equations from uploaded images in real time
Building an AI agent capable of conversationally discussing equations and concepts
Implementing retrieval-augmented generation with textbook grounding
Creating a fully voice-enabled tutoring experience
Building a phone-camera workflow with QR-based live capture
Persisting images directly inside MongoDB for stateless deployment
Designing an editorial-inspired educational interface with marginalia and retrieval traces
Creating a dynamic AI tutor that feels interactive instead of static