Agentic AI Scientist - Office of Data Science

St. Jude Children’s Research Hospital is where those with a passion for making a difference come to break new ground. Located in Memphis, Tennessee, St. Jude leads the way the world understands, treats, and defeats childhood cancer and other life-threatening diseases through groundbreaking research and compassionate care. 

The Office of Data Science (ODS; www.stjude.org/ods) at St. Jude institutionalizes and expands data science capabilities to unlock the vast potential of St. Jude’s biological data repositories. With a vision for St. Jude to become a global leader in applying data science to biological discovery, the ODS is creating a seamless ecosystem where experimental, clinical, and data scientists collaborate to accelerate discoveries, support scientific initiatives, and advance the understanding and treatment of pediatric catastrophic diseases.

The ODS is seeking a skilled and collaborative Agentic AI Scientist to lead the design, development, and evaluation of agentic AI systems that support research and operational workflows across the Office of Data Science. This role will focus on building and orchestrating advanced LLM-based agents, developing secure and interoperable infrastructure, and translating emerging AI capabilities into practical, high-impact solutions. The successful candidate will also contribute to technical reporting, results evaluation, and staff guidance to promote responsible and effective adoption of agentic AI. The position reports to the Chief of Artificial Intelligence.

Responsibilities

• Lead computationally focused agentic-AI research and engineering projects with moderate supervision from the manager; provide inputs on analytical approaches, set up experiments, and develop new agentic methods with manager input.
• Design and orchestrate agent systems (agent hierarchies, tool use, handoffs, checkpointing, and human-in-the-loop control) over ODS data and tools.
• Stand up and maintain MCP servers/clients and A2A-interoperable agents that connect ODS assets (e.g., St. Jude Cloud, internal warehouses, planning systems) under secure, auditable, HIPAA-aligned patterns.
• Identify, process, organize, summarize, review, and report results; instrument every agent with tracing, evaluations, and cost/latency budgets.
• Draft manuscripts and technical reports and work through submission and internal review with manager input; instruct and guide other staff in agentic-AI techniques.
• Mentor others in agentic AI best practices and standards 

Qualifications

• Minimum: Bachelor's in Computer Science, Bioinformatics, Biomedical Engineering, or a related field with 6+ years of relevant experience (OR) Master's with 4+ years (OR) entry-level PhD with no work experience. (PhD requirement waived, given an extraordinarily unusual level of scientific contribution.)
• Publication benchmark: 1 to 2 first-author papers with a journal impact factor > 5 (or equivalent contribution to other research / open-source outputs).
• Preferred: prior applied-ML or agentic-AI research; demonstrated shipping of LLM-integrated systems; healthcare, biomedical, or other research-intensive domain experience.
• Experience in project coordination, maintenance, and other research support activities preferred.

Preferred Technical / Functional Requirements (Agentic AI specialization):

Depth in at least one item per row; working awareness across the row. (Named technologies are illustrative examples, not a required stack.)

• Agent frameworks. Production depth in at least one framework (e.g., LangGraph, OpenAI Agents SDK, Google ADK, CrewAI, Microsoft Agent Framework, Pydantic AI, LlamaIndex, Strands, Haystack, DSPy, Agno).
• Interoperability protocols. Author and operate MCP (Model Context Protocol) servers, clients, and tools, including its core primitives (prompts, tools, resources) over JSON-RPC with OAuth 2.1 authorization; build A2A (Agent2Agent) interoperable agents; awareness of emerging agent standards (e.g., AG-UI, AGNTCY/OASF, MIT NANDA registry/discovery).
• Orchestration & control loop. The agent harness: model-output parsing, tool dispatch, context-window management, and enforcement of stop conditions, budgets, and safety boundaries; multi-agent topologies (e.g., hierarchy, swarm, graph); checkpointing and human-in-the-loop; composable, loadable agent capabilities (e.g., Agent Skills).
• Evaluation & observability. Tracing (e.g., OpenTelemetry), offline and online evaluation (e.g., LLM-as-judge), guardrails, and red-teaming, with observability tooling (e.g., LangSmith, Langfuse, Arize Phoenix, Microsoft Foundry Observ