Aditya Ramesh

Building autonomous AI agents that learn, reason, and act reliably in real environments.

About

Hi! I’m Aditya Ramesh. I work as a Research Scientist II at Fujitsu Research, India.

I build learning systems with an emphasis on turning learning algorithms into working decision-making systems that operate reliably under uncertainty in complex environments. I emphasize reliability ; safety and adversarial robustness, effective behavior in sparse-reward settings, and stability over extended operation in partially observable or changing environments ; and I typically carry projects end-to-end: identifying a research gap, formulating and publishing a method, and implementing it in a working proof-of-concept system.

You can find publications and technical details on my research page and Google Scholar. I also write longer technical notes and project writeups here.

Contact — Email / LinkedIn

News

  • “SPARK: Spectral Perturbation based Adversarial Attacks for KG-RAG Agents” accepted at ICLR 2026 Workshop (Agents in The Wild). Paper
  • Paper on improving performance in sparse reward settings : “Graph-theoretic intrinsic reward for RL” accepted at ICLR 2026 (in collaboration with Jatin et al.). Paper · Writeup
  • Paper on automating LLM jailbreaking : “Efficient Jailbreak attacks on LLMs”. Paper · Writeup

Timeline

Download CV (PDF)