Muntasir Adnan

I am an AI researcher and PhD candidate and my work sits at the intersection of automated software engineering, Large Language Model (LLM) robustness, and defensive cybersecurity. My research addresses critical failure modes in code generation models, specifically quantifying "debugging decay" and developing self-correcting feedback loops to raise accuracy and prevent Common Weakness Enumeration (CWE) security vulnerabilities. I am the creator of PyCapsule — a lightweight two-agent architecture for LLM self-debugging code generation and introduced the Debugging Decay Index (DDI), a quantitative framework predicting optimal intervention points in iterative AI debugging. My current work focuses on execution-guided evaluation harnesses that equip AI agents with interactive breakpoint debugging and Static Analysis Security Testing (SAST) guardrails to ensure robust, production-safe code generation.

Personal Website

https://github.com/Adnan525

LinkedIn

LinkedIn

Projects

No grants recorded.