You'll work with Python, PyTorch, AWS, Ray, PyTorch Lightning, and Weights & Biases to design and train machine learning models for Entrust's digital identity verification platform (formerly Onfido), tackling deepfake detection, bias mitigation, document understanding, and anomaly detection. The Applied Science team has approximately twenty machine learning scientists working with product engineering to deploy models for identity verification and facial biometrics at global scale.

Thomson Reuters★3.9 (4.3k)· London

You'll work with PyTorch, JAX, HuggingFace Transformers, vLLM, LLM APIs, Ray, Spark, and experiment tracking tools like Weights & Biases and MLflow on Thomson Reuters' Foundational Research division, which develops advanced algorithms and training techniques for Large Language Models including LLM training, post-training techniques for reasoning and workflows, data-centric ML, and evaluation. You'll design and implement scalable systems for training, fine-tuning, and evaluating LLMs while translating cutting-edge research into production implementations.

You'll lead a 6-person team of applied scientists building vision-language models for document data extraction and efficient mobile models (<5MB) using PyTorch, with ~50% hands-on coding time. The work spans vision-language modeling, document understanding, distillation, quantization, and few-shot learning to power Entrust's identity verification platform that detects fraud and deepfakes at global scale. You'll define roadmap with product/engineering, contribute code, and publish research while managing team development in a product-driven ML environment.