Responsibilities: Design, train, fine-tune, and evaluate ML models (LLMs, classification, sequence models) for property insurance. Build and maintain robust data pipelines that feed training, evaluation, and inference workloads at scale. Develop rigorous evaluation frameworks — establish metrics, build rater alignment processes, and apply statistical methods to determine when a candidate model is genuinely better than production. Run controlled experiments, ablations, and A/B tests; communicate findings clearly with appropriate uncertainty quantification. Deploy models to production and own their performance, drift monitoring, and iteration cycles. Collaborate with the engineering team to integrate ML services into our backend (Django/Python) and frontend (React/TypeScript) products. Stay current with the ML literature and translate relevant advances into practical improvements for our products.
Required:
PhD in Statistics, Machine Learning, Computer Science, Applied Mathematics, or a closely related quantitative field (or equivalent research experience with a strong publication or production track record). Strong foundation in statistics — experimental design, hypothesis testing, Bayesian methods, and uncertainty quantification. Minimum 5 years of combined research and applied ML experience, with a proven track record of shipping models to production. Deep proficiency in Python and the modern ML stack (PyTorch, Hugging Face, scikit-learn, pandas, NumPy). Hands-on experience with LLMs, including fine-tuning (LoRA/QLoRA, full fine-tunes), prompt engineering, and evaluation. Strong data engineering skills — comfort building reliable pipelines over messy real-world data, working with SQL and columnar formats. Excellent debugging, problem-solving, and written communication skills.