Self-Debugging C/C++ CodeGen
During this winter break I attended a 4-day bootcamp held at my university by a pair of InvoZone employees. Here, I learned some of the fundamentals of Applied AI, specifically n8n Automations and Agentic AI. This was my capstone for the bootcamp.
Crunner works by providing LLMs 3 simple tools:
- the ability to create files and folders in a specified folder
- the ability to call a function that runs 'make' in a subshell and returns the output and/or error messages
- the ability to call a function that runs the compiled code in a subshell and returns the output and/or error messages
This is achieved by rerouting stdout and stderr for the subprocess to a temporary file that is later read-from to provide the LLM feedback. This generation-feedback loop continues until a satisfactory answer has been reached.