Cheng-Yuan (Ross) King

I'm an MSc Artificial Intelligence student at the University of Sheffield, with a Computer Science background from Queen's University Belfast. My focus is empirical AI safety and evaluation — designing and running experiments that measure model behaviour, failure modes, robustness, and safe refusal reliability.

Recent work centres on AI safety and evaluation: a synthetic internal AI agent evaluation lab covering RAG retrieval benchmarks, structured extraction, red-team safety testing, safe refusal evaluation, safety classifiers, tool governance, OTel-style observability, and a public Streamlit dashboard (350 golden cases, 60 red-team cases, release gate reporting); an LLM red-team harness measuring attack success rate across 12 evaluation cells with cross-judge validation and bootstrap confidence intervals; and an LLM event extraction baseline comparing unconstrained vs. constrained-label prompting on MAVEN and WikiEvents with Qwen2.5-7B on A100 GPU. Supporting production-systems evidence: a synthetic fintech analytics platform with a dbt/BigQuery warehouse, activation model, CUPED experimentation, and a GCP/Cloud Run deployment path.

I prefer building end-to-end systems over isolated notebooks. I care about reproducibility, honest evaluation, and clear documentation of limitations — understanding failure modes as clearly as the wins. When something scores well, I want to know whether the benchmark is actually well-posed, which is why cross-judge validation, benchmark transparency, and guardrail metrics show up across most of my projects.

I'm currently looking for empirical AI safety research, AI evaluation, and applied AI/ML roles in the UK from October 2026, with particular interest in measuring model behaviour, robustness, and safety. I have a UK Graduate Visa route lined up, so no sponsorship is needed for two years post-graduation.

The fastest way to reach me is by email, or through LinkedIn.