Senior Data Scientist at Refractal
United Kingdom
Posted
on-site · full_time · Visa sponsorship
Apply for this role at Refractal
Responsibilities:
- Own production detection models that make millisecond safety decisions (train, tune, deploy).
- Design low-latency architectures (distilled/small models, rules+ML ensembles) and optimise inference cost/latency.
- Analyse telemetry to detect drift, blind spots and new attack patterns; feed findings into model improvements.
- Build robust evaluations, benchmarks and labelling processes; collaborate with security engineers to convert attacks into training data.
- Instrument per-customer quality tracking and translate policies/regulatory requirements into measurable controls.
Requirements:
- 5+ years applied ML/data science experience with production deployments.
- Strong Python and PyTorch (or equivalent) skills and solid statistical evaluation ability.
- Hands-on NLP/LLM experience (fine-tuning, embeddings, classification, LLM-as-judge workflows).
- Experience with labelling, inter-annotator agreement, drift detection and trade-off communication.
- Ability to explain technical trade-offs to engineering and customer teams.
- Nice-to-have: experience in adversarial ML, safety/content-moderation models, and published evaluations or AI security research.