EasyMat — AI Math Tutor for the Israeli Bagrut

An AI-powered math tutor for Israeli high school students preparing for the Bagrut exam (3, 4, and 5 units). Combines RAG architecture with the Claude API to deliver context-aware, curriculum-aligned answers with full LaTeX rendering.

Features

• RAG pipeline: Obsidian-style markdown notes -> Voyage AI embeddings -> Pinecone vector search
• Streamed Claude API responses with prompt caching (~90% cost reduction)
• LaTeX rendering via KaTeX for clean formula display
• Image upload for handwritten equations - Claude transcribes to LaTeX and solves
• Unit level selector - adapts tone and curriculum to 3, 4, or 5 units
• Bilingual Hebrew/English with automatic RTL/LTR switching

Tech Stack

Next.js 14 · TypeScript · Claude API (Anthropic) · Pinecone · Voyage AI · KaTeX · Tailwind CSS

Getting Started

Prerequisites

Three API keys are required:

• Anthropic - Claude API
• Pinecone - vector database (free tier)
• Voyage AI - embeddings (free tier)

Setup

git clone https://github.com/SaharHalili95/easymat.git
cd easymat
npm install

Create .env.local:

ANTHROPIC_API_KEY=your-key
PINECONE_API_KEY=your-key
PINECONE_INDEX_NAME=bagrut-notes
VOYAGE_API_KEY=your-key
INGEST_SECRET=any-random-string

npm run dev

Ingest Notes into Pinecone

curl -X POST http://localhost:3000/api/ingest \
  -H "x-ingest-secret: your-ingest-secret"

Project Structure

src/
  app/
    page.tsx              # Chat UI
    api/chat/route.ts     # Claude + RAG endpoint (streaming)
    api/ingest/route.ts   # Notes ingestion endpoint
  components/
    ChatMessage.tsx       # Renders messages with LaTeX
    ChatInput.tsx         # Text input + image upload
    UnitSelector.tsx      # 3/4/5 unit level picker
  lib/
    claude.ts             # Anthropic client singleton
    pinecone.ts           # Pinecone client singleton
    rag.ts                # Vector retrieval
    embed.ts              # Voyage AI embeddings
    buildSystemPrompt.ts  # RAG context + system prompt
  data/notes/             # Markdown knowledge base (Bagrut curriculum)

GitHub

github.com/SaharHalili95/easymat