Principal Data Scientist

Role overview Build production NLP and applied AI systems that turn large collections of scientific articles into structured research intelligence. You’ll own domain-specific modeling work from evaluation design through deployment, balancing model quality, operating cost, and reliability while working directly with a small, experienced team and product stakeholders.

Responsibilities - Develop scalable pipelines that extract entities, classifications, claims, and summaries from scientific full text. - Compare classical NLP, embeddings and retrieval, LLM prompting, and smaller fine-tuned models; select methods based on measured quality, cost, and operational needs. - Create evaluation datasets with subject matter experts and vendors, define metrics, and use results to guide model improvements. - Write production Python and manage concurrency, reliability, and cost for high-volume model workloads. - Build pipeline stages that can be retried, evaluated, and safely rerun; coordinate orchestration with data engineering colleagues. - Contribute modeling and evaluation expertise to tool-using AI applications, and translate feedback from editors and product partners into concrete improvements.

Requirements - Deep experience writing and maintaining production Python systems at scale, including sound choices among asynchronous, threaded, and queued workloads. - Strong knowledge of both modern NLP methods—such as transformers, LLMs, embeddings, and retrieval—and classical methods such as named-entity recognition, classification, and sequence labeling. - Experience designing evaluations, interpreting results, and addressing model performance changes over time. - Ability to compare approaches rigorously and justify choices with quality measurements and cost estimates. - A record of shipping models or data products that deliver value to users, beyond prototypes or research work.

Nice to have - Experience with scientific or scholarly text, production LLM workloads under cost and latency limits, or agentic systems with tool use and safeguards. - Familiarity with AWS services such as S3, Batch, Lambda, and SageMaker, and data lake formats or patterns such as Parquet and Iceberg.

Benefits and work setup - The source describes a comprehensive benefits package and meeting-free Friday afternoons for focused work and development. Compensation varies by location and candidate qualifications.