Program Aided Math Problem Solving Agent with Python and SymPy

Does a small local language model solve maths problems better when it writes programs? This project builds a Haystack agent with a sandboxed Python tool and SymPy solve and verify tools; a program runs in a temporary folder, with no network and a time limit, and errors go back to the model. The model (qwen2.5:1.5b) runs in local Ollama on a laptop with RTX 2050 4 GB (Ryzen 7, 8 GB RAM).

NOTE: Research / teaching demo on the public GSM8K and MATH-500 benchmarks. A small local model can be wrong: check the program and the result before relying on an answer. Programs run only in a temporary folder, without network access and with a time limit.

Category: Department of Computer Science and Engineering / Agent Development

Agent Development Project AG16 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.