Lead Applied Scientist, Document Understanding

Lead Applied Scientist, Document Understanding About the Role: This role sits within the applied science function and owns the design, development, and production deployment of document understanding systems that power Westlaw, PracticalLaw, and CoCounsel. You will work on semantic chunking, document enrichment, knowledge graph construction, and synthetic data generation for complex legal, tax, and accounting content. Multiple product teams depend on this function. What You'll Do Design and deploy semantic chunking models for lengthy, non‑uniformly structured legal documents with adjustable granularity across use cases. Build document enrichment systems using legal and customer-defined taxonomies. Develop LLM‑based knowledge graph construction pipelines that extract and link citations, entities, and legal concepts across diverse legal content. Lead knowledge distillation efforts to compress large models into latency‑constrained, production‑ready SLMs. Design evaluation frameworks – component‑level and end‑to‑end – using expert annotation and synthetic data. Own technical decisions on architecture, chunking strategy, classification approach, and knowledge extraction methods. Partner with engineering on delivery, reliability, and scale across multiple product lines. Provide technical input to senior leadership on AI strategy and roadmap. Mentor applied scientists and ML practitioners on the team. Required Qualifications PhD in Computer Science, AI, NLP, or a related field – required. 8+ years of post‑degree industry experience shipping document understanding, information extraction, or knowledge graph systems into production. Publications at ACL, EMNLP, ICLR, NeurIPS, SIGIR, KDD, or equivalent. Production Python and experience with PyTorch, Hugging Face Transformers, and DeepSpeed. Hands‑on production depth in: Document layout analysis and semantic chunking beyond fixed‑size or paragraph‑based methods. Hierarchical, multi‑label document classification with domain‑specific and customer‑defined schemas. Entity recognition and linking, relation extraction, citation parsing, and knowledge graph construction from unstructured text. LLM‑based information extraction, few‑shot and multi‑task learning, and post‑training. Knowledge distillation, model compression, and SLM deployment under latency constraints. Synthetic data generation and annotation workflow design. End‑to‑end evaluation framework design for document understanding. Preferred Qualifications Legal document understanding, legal IE, or legal AI experience. Experience with complex document structures: nested hierarchies, cross‑references, non‑uniform formatting. Retrieval or QA systems over large document collections. RAG and agentic workflows in enterprise settings. Knowledge graph frameworks for legal or enterprise applications. Experience with AzureML or AWS SageMaker. Compensation & Benefits Base compensation range (US locations): $147,600 USD – $274,200 USD. Ontario, Canada: $140,000 CAD – $175,000 CAD. Compensation is positioned based on experience, skills, and internal equity. The role may also be eligible for an annual bonus based on enterprise and individual performance. Benefits include flexible vacation, two company‑wide mental health days off, access to the Headspace app, retirement savings, tuition reimbursement, employee incentive programs, and resources for mental, physical, and financial wellbeing. Other benefits: optional hospital and accident insurance paid 100% by the employee; optional life and AD&D insurance paid 100% by the employee; flexible spending and health savings accounts; fitness reimbursement; employee assistance program; group legal identity theft protection; 529 plan; commuter benefits; adoption and surrogacy assistance; employee stock purchase plan. Equal Opportunity and Accommodations Thomson Reuters is a proud Equal Employment Opportunity Employer. We provide reasonable accommodations for qualified individuals with disabilities and for sincerely held religious beliefs in accordance with applicable law. #J-18808-Ljbffr Refinitiv