Project: AI Math Tutor

Dec 30, 2025 • 30 min read

Standard LLMs are terrible at math. They "hallucinate" numbers because they predict the next token instead of calculating the result. Asking GPT-4 to multiply 6-digit numbers is like asking a poet to do accounting.

In this guide, we will build a Math Tutor Agent that uses OpenAI's Code Interpreter to write and execute Python code for 100% accurate calculations.

1. The Architecture: "Don't Compute, Orchestrate"

2. Setting up the Assistant with Code Interpreter

We only need to do this once. The Assistant lives on OpenAI's servers.

from openai import OpenAI

client = OpenAI(api_key="sk-...")

assistant = client.beta.assistants.create(
    name="Professor Pi",
    instructions="You are a math tutor. Always write Python code to solve problems. Use SymPy for algebra and Matplotlib for graphs.",
    tools=[{"type": "code_interpreter"}], 
    model="gpt-4-turbo"
)

print(f"Assistant ID: {assistant.id}")

3. Solving Symbolic Math (Algebra)

Most tutorials just show `1 + 1`. Let's solve Algebra.

The Assistant will automatically generate and run this:

import sympy
x = sympy.symbols('x')
expr = sympy.sin(x**2)
derivative = sympy.diff(expr, x)
print(derivative)

Output: 2*x*cos(x**2)

4. Generating Graphs (Visual Learning)

A good tutor draws diagrams. Code Interpreter can generate image files.

import matplotlib.pyplot as plt
import numpy as np

x = np.linspace(-10, 10, 100)
y = x**3

plt.plot(x, y)
plt.title("y = x^3")
plt.grid(True)
plt.savefig('/mnt/data/plot.png') # <--- Saves to sandbox

5. Handling Image Files in the Frontend

When the run completes, you need to check `messages` for image files.

// Node.js: Retrieve the file content
const messages = await openai.beta.threads.messages.list(thread.id);
const lastMsg = messages.data[0];

if (lastMsg.content[0].type === 'image_file') {
  const fileId = lastMsg.content[0].image_file.file_id;
  
  // Download the raw bytes
  const response = await openai.files.content(fileId);
  const imageBuffer = await response.arrayBuffer();
  
  // Send to frontend as Base64 or Blob
}

6. Integrating LaTeX Rendering

For the text explanations, tell the model to output LaTeX (e.g., $2x^2$). Then use a library like react-katex to render it beautifully.

System Instruction Tweak

"Output all mathematical expressions in LaTeX format wrapped in single dollar signs."

import 'katex/dist/katex.min.css';
import Latex from 'react-latex-next';

// In your chat component
<div className="message-bubble">
  <Latex>{message.content}</Latex>
</div>

7. Next Steps

This is just the beginning. The Assistants API also supports Function Calling (connecting to external APIs like Wolfram Alpha) and File Search (uploading textbooks as knowledge).