Samuel Nellessen

Bachelor Student in Artificial Intelligence
Radboud University Nijmegen, Netherlands
📍 Nijmegen, Netherlands

✍️

Composing haiku...

About Me

I am an AI Safety Researcher building systems that autonomously find failure modes in Large Language Models using Reinforcement Learning.

I pivoted from Cognitive Science and Philosophy to technical AI because I believe beliefs should pay rent in concrete results. Now, I apply that rigor to engineering verifiable safety frameworks.
Download CV

Current Work

I am currently a researcher at KachmanLab developing automated jailbreaking frameworks. My work focuses on:

• Reinforcement Learning: Implementing verifiable reward frameworks (using GRPO and Verifiers) to train adversarial agents.
• Inference Optimization: Designing asynchronous generation pipelines using asyncio and vLLM to maximize throughput across multi-GPU environments.
• Evaluation: Benchmarking model robustness against steganographic reasoning and alignment faking.

Background

• ARENA 5.0 Fellow: Selected for the 2025 cohort to specialize in LLM reasoning and mechanistic interpretability.
• Foresight Fellow: Advising for technical grants and researched computational models of agency.
• Donders Institute: Built Bayesian models for computational psychiatry (controllability) with Roshan Cools.

News

Education

Publications

Blog

Projects