AlgebrAI: A Code-Generating Linear Algebra Calculator

AlgebrAI is an AI-powered calculator built to solve linear algebra problems with high precision.

Unlike general-purpose language models that often produce incorrect or inconsistent math outputs, AlgebrAI generates and executes reliable Python code to directly compute the accurate final answer.

It is powered by a fine-tuned Qwen 3 1.7B model enhanced with QLoRA adapters, and trained on a custom dataset of linear algebra problems paired with executable Python solutions. When a user inputs a natural language math problem, the model:

1. Understands the query,
2. Generates valid Python code to solve it (e.g., SVD, eigenvalues),
3. Executes the code, and
4. Returns the final result.

🧠 What Makes This Different

• Purpose-built for linear algebra

Unlike generic LLMs that fumble math, AlgebrAI is fine-tuned specifically to handle linear algebra queries with precision and consistency.

• Direct code execution

It converts math queries into executable Python code, and returns only the final answer with no fake steps or guesswork.

• Custom Training Data

Trained on a hand-crafted dataset of 6000+ real linear algebra problems.

📂 Project Files

 ├── linear_algebra.ipynb # Main notebook (run your queries here) 
 ├── final_dataset_no_comments.json # Training/inference dataset
 └── README.md # Project documentation 

---

🚀 Run It Yourself

1. Clone the Repo

 
   git clone https://github.com/Umme-2003/AlgrbAI.git 
   cd AlgrbAI 

2.Install dependencies

   pip install numpy torch transformers peft accelerate

⚠️ GPU recommended for faster model inference

3. Launch the notebook

   jupyter notebook

Open linear_algebra.ipynb and run the cells step-by-step. You can enter your own linear algebra problem and watch the model solve it with generated Python code.

✨ Sample Usage Input Prompt:

   Let A = [[1, 2], [3, 4], [5, 6]]. Perform Singular Value Decomposition (SVD) on matrix A and return its singular values.

Model Output:

import numpy as np
A = np.array([[1, 2], [3, 4], [5, 6]])
U, S, Vt = np.linalg.svd(A)
print(S)

Final Answer:

   [9.5255, 1.5143]

---

🧠 Model Info

• Base LLM: Qwen 3 1.7B

• Tuning: LoRA adapters on a curated problem-code dataset

• Inference Tools: Hugging Face Transformers + PEFT

• Execution Env: Jupyter / Lightning AI

---

Use Cases

• 🧑‍🎓 Students verifying linear algebra homework
• 👩‍💻 Developers testing numerical algorithms
• 🤖 Researchers benchmarking LLMs for mathematical reasoning