Verified Math Problem Solver

A hybrid mathematics application in which SymPy performs the calculation and verification while an optional language model explains the already-verified result.

This is intentionally different from a general chatbot: the LLM is not trusted to calculate the final answer.

Logic

flowchart LR
    A["Equation or expression"] --> B["Restricted parser"]
    B --> C["SymPy solver"]
    C --> D["Residual verification"]
    D --> E["Verified answer"]
    E --> F["Optional LLM explanation"]

Loading

Supported tasks

• arithmetic and simplification
• single- and multi-variable symbolic equations
• implicit multiplication such as 2x
• exponent notation with ^
• substitution-based residual checks

Run

git clone https://github.com/sandeep848/langchain-math-problem-solver.git
cd langchain-math-problem-solver
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run app.py

The Groq key is optional because solving and verification run locally.

Test

pytest

Design boundary

Natural-language word problems still require an interpretation layer. The repository deliberately keeps that separate from symbolic computation so that a fluent explanation cannot silently replace mathematical verification.