Lead Scientist, Document Understanding & AI Pipelines

Thomson Reuters Corp.; New York (NY); On-site; USD 148,000 - 274,000; Full time

Job summary

Thomson Reuters Corp. in New York, NY is seeking a Lead Applied Scientist to design, develop, and deploy document understanding systems powering Westlaw, PracticalLaw, and CoCounsel.

You will tackle semantic chunking, document enrichment, knowledge graphs, and synthetic data for complex legal content. The role requires a PhD, 8+ years in industry, and hands-on experience with PyTorch, Transformers, and production ML pipelines.

Qualifications

• PhD in Computer Science, AI, NLP, or related field.
• 8+ years in industry building document understanding or NLP.
• End-to-end ML lifecycle including production deployment.
• Hands-on PyTorch, Transformers and DeepSpeed experience.
• Publications in ACL/EMNLP/NeurIPS or equivalent.

Responsibilities

• Design and deploy semantic chunking for long legal documents.
• Build document enrichment systems using legal taxonomies.
• Develop LLM-based knowledge graph construction pipelines.
• Lead knowledge distillation for latency-constrained production SLMS.
• Design evaluation frameworks with expert annotation and synthetic data.
• Collaborate with engineering on reliability and scale across products.
• Mentor applied scientists and ML practitioners on the team.
• Contribute to AI strategy and roadmap discussions.

Skills

PhD in CS/AI/NLP

8+ years experience

Python (PyTorch)

Hugging Face Transformers

DeepSpeed

Mentoring

Education

PhD in CS/AI/NLP

Tools

PyTorch

Hugging Face Transformers

DeepSpeed

Job description

Thomson Reuters Corp. in New York, NY is seeking a Lead Applied Scientist to design, develop, and deploy document understanding systems powering Westlaw, PracticalLaw, and CoCounsel.

You will tackle semantic chunking, document enrichment, knowledge graphs, and synthetic data for complex legal content. The role requires a PhD, 8+ years in industry, and hands-on experience with PyTorch, Transformers, and production ML pipelines.