Agentic C++ Dry Run Code Generator and Runner

A tool for generating and testing "dry runs" (tricky C++ code snippets often asked in exams by universities like FAST) in an isolated Docker environment.

Overview

This project provides a system for:

• Generating challenging C++ code snippets that test understanding of tricky language concepts taking inspiration from various PF and OOP past papers through Retrieval Augmented Generation (RAG).
• Running the code snippets safely in a virtual execution environment using Docker, to ensure accuracy.
• Automated testing and verification (as the outputs are fed back to a reasoning LLM, which generates until it has 10 various questions that meet the criteria and are worth studying).
• A web interface to view the newly generated code snippets, their outputs, and an explanation for the output.

Features

• Focuses on tricky C++ concepts like:
  • sizeof() and datatype size edge cases
  • Order of operations (e.g., x++ + x++)
  • Conditional statement edge cases
  • Loop patterns
  • Pointer manipulation and const correctness among many others asked by FAST in courses like Programming Fundamentals (PF) and Object Oriented Programming (OOP).

Setup

1. Prerequisites:

   • Python 3.x
   • Docker
   • Required Python packages (see requirements)

2. Environment Variables:

   • Generate an API Key from Google AI Studio
   • Create a .env file with:

     GOOGLE_API_KEY=your_api_key_here
     

Usage

With python virtual env:

1. Download install.sh
2. Set it as executable: chmod +x install.sh
3. Wait.
4. When done, set run.sh as executable: chmod +x ./run.sh
5. Run it. ./run.sh

Alternatively, for those with conda:

1. Download install_condabased.sh
2. Follow the steps above.