Ludivine Siau
Operations & Programme Delivery in AI Safety
Fourteen years of experience delivering complex digital programmes across government and SaaS scale-ups, paired with graduate research at UCL in causal machine learning for bias mitigation in clinical AI.
Focused on operationalising AI governance, helping AI safety organisations scale their impact, and bridging the divide between technical research, policy-making, and civic engagement.
Operational Delivery & Scaling
Fourteen years of experience getting things built and delivered across UK government and tech scale-ups. I help teams cut through ambiguity, figure out what actually needs doing, and execute under pressure:
- Sensible governance: Putting practical processes, oversight, and decision-making structures in place where none existed.
- Cross-discipline collaboration: Aligning engineers, policy teams, domain experts, and external suppliers to launch complex public infrastructure.
- Unblocking execution: Fixing practical bottlenecks across disciplines (e.g. hiring, compliance) and ruthlessly prioritising effort under tight deadlines.
AI Safety, Ethics & Society
I write to bridge the gaps between three groups that often talk past each other: the technical AI safety community, policymakers, and the public. My focus is on grounding abstract ethical debates in technical and operational reality:
- Technical risks & transparent evaluation: Unpacking the real-world implications of technical AI risks, safety benchmarks, and what genuinely transparent development and evaluation look like in practice.
- Human agency & societal resilience: Exploring how AI deployment affects individual agency, democratic structures, and societal resilience against systemic harms.
- Applied AI ethics: Moving beyond vague principles to examine the hard trade-offs between capability, safety, and genuine public benefit.
AI Fairness Research
Graduate research at UCL focused on causal machine learning to detect and mitigate bias in clinical AI.
- Developed a Structural Causal Model and the CEVAE-HE framework to separate genuine biological variation from unfair demographic bias.
- Built an agentic pipeline to generate synthetic data for fairness research and stress-testing.