Bio

I am a final-year MSCA Doctoral Fellow at Imperial College London, advised by Prof. Alessandra Russo and Prof. Joey Bose. My research interests focus on test-time adaptation and post-training for generative foundation models to advance automated scientific research and embodied AI.

I did a research internship at Microsoft Research AI Frontiers in New York, hosted by Siddhartha Sen and John Langford. Previously, I collaborated with Google DeepMind on benchmarking LLMs on hard STEM tasks. I also worked at Microsoft AI on post-training Copilot 365 for tool use.

Before pursuing a PhD in AI, I co-founded Athina AI (YC W23), an LLM observability startup. I also hold an MSc in Data Science from TU Munich, where my thesis was supervised by Prof. Stephan Günnemann and hosted at BMW.

I co-organize the ICARL Seminar at Imperial and am a mentor at SPAR. Please feel free to reach out if you share similar research interests or would like to discuss transitioning from the startup world to a PhD.

Selected Publications & Preprints

Fork-and-Flush: Escaping Idea Basins in Autoresearch Agents.
Z. Cai*, C. Ziakas*, V. Kontonis, T. Pearce, S. Sen, A. Krishnamurthy, S. Garg, D. Papailiopoulos.
Preprint (under review), 2026.

Aligning Flow Map Policies with Optimal Q-Guidance.
C. Ziakas*, A. Russo, A. J. Bose.
NeurIPS 2026.

Grounding Generated Videos in Feasible Plans via World Models.
C. Ziakas*, A. Bar, A. Russo.
Preprint (under review), 2026.

VITA: Zero-Shot Value Functions via Test-Time Adaptation of Vision-Language Models.
C. Ziakas*, A. Russo.
ICLR 2026.

Red-Bandit: Test-Time Adaptation for LLM Red-Teaming via Bandit-Guided LoRA Experts.
C. Ziakas*, N. Loo, N. Jain, A. Russo.
ACL 2026.

For a full list, please visit Publications & Preprints page.

* indicates equal first authorship.