Qwen-GeoGebra-Coder-7B

<p align="center"> <img src="logo.png" alt="GeoGebra AI Logo" width="550"/> </p>

Qwen-GeoGebra-Coder-7B

Developed by: Khurram Pervez, Assistant Professor of Mathematics

This repository contains a specialized AI pipeline designed to transform natural language mathematical descriptions into real-time GeoGebra 3D visualizations. By combining a fine-tuned Qwen-Coder backbone with a custom FastAPI processing layer, the system bridges the gap between abstract reasoning and geometric plotting.

🚀 How It Works

The system operates via a three-stage pipeline designed for mathematical accuracy:

1. Chain-of-Thought (CoT) Reasoning

When a user submits a prompt (e.g., "Create a cylinder with radius 3 and height 10"), the model first enters a <thought> state. It calculates the spatial coordinates, base-point, and top-point vectors required by GeoGebra's engine before generating any code.

2. Smart Command Extraction (The FastAPI Layer)

LLMs often output coordinates in various styles (e.g., [<0,0,0>] or [3,0,0]). GeoGebra is strict about its syntax: Cylinder(Point, Point, Radius).

Our custom clean_and_format_ggb function acts as a mathematical transpiler:

• Standardization: Converts angle brackets < > and square brackets [ ] into standard parentheses ( ).
• Coordinate Mapping: Identifies base/top points and extracts the scalar radius from the spatial reasoning.
• Validation: Ensures the final string is a valid GeoGebra command like Cylinder((0,0,0), (0,10,0), 3.0).

3. Real-Time Rendering

The app.html interface uses the GeoGebra Discovery/Classic API to inject the generated commands into a live 3D canvas, allowing for immediate visual verification of mathematical concepts.

📸 Interface Example

<p align="center"> <img src="example.png" alt="GeoGebra AI Research Interface" width="100%"/> </p>

🛠️ Installation & Setup

Prerequisites

• Hardware: NVIDIA RTX 4060 Ti (16GB VRAM recommended).
• Environment: Ubuntu 22.04+, Python 3.10+, llama-cpp-python.

Local Deployment

To run this research interface on your local machine:

1. Download Files: Download server.py, app.html, and math_viz_Q4_K_M.gguf into one folder.
2. Install Requirements:

   pip install fastapi uvicorn llama-cpp-python