Debangan Mishra

Conducting research on AI Safety, particularly machine unlearning and adversarial attacks against ML models. Contributed to the creation of a SOTA graph unlearning method capable of erasing poisons in a corrective setting. Benchmarking the robustness of multimodal LLMs against adversarial attacks and jailbreaks, particularly white-box attacks.

Worked on a multi-agent POC using AutoGen to verify Infrastructure as Code (IaC) on the Azure Cloud Platform against the Azure Security Benchmark (ASB), enabling pre-deployment compliance checks for developers and built a VSCode extension to interface with it

Developed a high-performance B2B e-commerce website, integrating Zoho REST APIs. Utilized Next.js and React to create a fast and efficient web application.