Data Scientist - Agentic AI

About the role

Hewlett Packard Enterprise is hiring a Data Scientist - Agentic AI in San Juan, Puerto Rico. The role builds and operationalizes the core agentic workflows behind Marvis, a next-generation AI assistant for network operations. It sits at the intersection of data science, generative AI and production engineering, and is designed as hybrid, with about two days per week in an HPE office.

Key responsibilities

• Design, implement and iterate agentic workflows using LangGraph, including ReACT loops, dynamic tool selection, multi-step reasoning and self-correction.
• Build and maintain MCP servers and skills: tool schemas, domain tools, skill playbooks and server lifecycle management.
• Integrate LLM capabilities at production scale, including structured outputs, streaming, function calling, prompt engineering and error handling.
• Build retrieval and memory services such as RAG pipelines, vector-store search, hybrid retrieval and long-term agent memory.
• Design evaluation frameworks for non-deterministic agents, including metrics, test harnesses, A/B tests and continuous quality improvement.
• Work with network engineers and product managers to turn troubleshooting playbooks into skills, tools and data pipelines.
• Develop data analysis and transformation logic in sandboxed execution environments, including multi-tool orchestration and visualization.
• Deploy containerized services in Kubernetes, contribute to CI/CD pipelines and manage container images.
• Own observability for agentic workflows: tracing, logging, cost tracking and performance monitoring.

Required profile

• Bachelor's, Master's or PhD in computer science, data science, mathematics, statistics or a closely related quantitative discipline.
• Typically 4-7 years of experience building production ML/AI systems, with at least a year of hands-on work with generative AI and LLM applications.
• Subject-matter expert in agentic AI and applied GenAI, able to lead projects and mentor colleagues.

Required skills

• Agentic orchestration frameworks: LangGraph strongly preferred, also LangChain, Claude Agent SDK or equivalent.
• MCP or equivalent tool-serving protocols, and LLM API integration at scale.
• RAG pipelines, vector stores (OpenSearch or equivalent) and evaluation frameworks (A/B testing, LLM-as-judge).
• Python in production codebases, not just notebooks.
• Kubernetes, CI/CD (Jenkins, GitHub Actions, ArgoCD), Docker, FastAPI or equivalent, and observability tools such as LangFuse or Prometheus.

Preferred: agent memory systems, sandboxed code execution (E2B, Firecracker), networking domain knowledge, AWS Bedrock or OpenSearch Serverless, and strong communication skills. Benefits include health and wellbeing support, professional development programs and an inclusive workplace.