Senior Data Scientist, Agentic AI & LLM Engineering

Description CAS uses unparalleled scientific content, specialized technology and unmatched human expertise to help R&D organizations across Commercial, Government and Academic sectors create groundbreaking innovations that benefit the world. As the Scientific Information Solutions Division of the American Chemical Society, CAS manages the largest curated reservoir of scientific knowledge, and for 119 years, has helped innovators mine, assess and apply that information to keep businesses thriving. The CAS team is global, diverse, endlessly curious and strives to make actionable scientific insights accessible to innovators worldwide. CAS is currently seeking a Senior Data Scientist focused on agentic AI development and large language model (LLM) engineering. This position will be located at our headquarters in Columbus, Ohio. This role sits on the Data Analytics and Insights (DAI) team, which builds the AI that powers CAS's scientific information products. You will work on the systems behind the Newton research assistants in SciFinder and BioFinder, the natural-language query agent in IPFinder, predictive models that ship inside products, and AI-assisted content curation pipelines, alongside data engineers, product managers, and scientific domain experts. This role requires a highly self-directed professional who can independently design, build, and evaluate agentic AI systems and LLM-powered applications that turn CAS's curated, provenanced scientific content into trustworthy research capabilities. The successful candidate will provide strategic input into product and platform ideation, drive complex technical initiatives from concept through production, and translate advanced AI engineering into tangible business value with minimal oversight, while providing technical direction to less-experienced scientists and engineers. Key Responsibilities Agentic AI & LLM System Development - Independently design, build, and productionize agentic AI systems - including multi-step, tool-using, and multi-agent workflows - that support scientific research tasks. - Develop retrieval-augmented generation (RAG) pipelines that ground model outputs in curated, provenanced scientific content. - Implement and extend model-context and tool-integration frameworks (for example, the Model Context Protocol) to connect LLMs and agents to scientific data sources and external systems. - Apply prompt engineering, context engineering, and orchestration techniques to improve the reliability and quality of agentic and generative workflows. Applied AI Engineering & Model Adaptation Fine-tune, adapt, and deploy large language models - including self-hosted, open-weight models - for domain-specific scientific tasks. Develop and deploy machine learning models that ship inside CAS products, such as property, toxicity, or biologic developability prediction, using scientific judgment to inform feature design and validate model behavior. Take AI capabilities from prototype all the way through production: building the data, training, and inference pipelines, owning the deployment, and iterating after first release, not just the proof of concept. Work with engineering and product to put AI features in front of real users at scale, and stay accountable for how they behave once they're live. Set and hold the engineering standard for production AI: appropriate safeguards such as content-safety guardrails, entitlement-based access control, evaluation, and model fallback, delivered as clean, tested, well-documented software with unit, integration, and end-to-end tests, containerized development, and CI/CD. Use agentic software development tools such as Claude Code as a core part of the daily workflow, and set the team's practice for usin