Government Incubator for Artificial Intelligence (i.AI)
Technical Program Manager, Safety Evaluations
No. 10-founded AI start-up transforming UK public services
- Lead the Sovereign Education Benchmark, the UK's first national safety evaluation of AI tutoring systems used by children — owned end to end, from pitching the Prime Minister's AI Adviser to technical delivery.
- Built a novel harm taxonomy for child-AI interactions grounded in KCSIE statutory guidance and the Online Safety Act, translating safeguarding frameworks into evaluation criteria a machine-learning team can implement.
- Designed and ran adversarial evaluations using Anthropic's Petri and Bloom frameworks and AISI's Inspect — thousands of synthetic multi-turn scenarios across child-safety risks — and built an LLM-as-judge adjudication pipeline calibrated against 100+ expert human annotators.
- Work across the AI Safety Institute, the Department for Education, frontier labs (Anthropic, Google DeepMind) and civil society (NSPCC, Internet Watch Foundation, 5Rights, the Children's Commissioner's Office).