HPE
Data scientist - Agentic AI
$156K–$315K
San Jose, California, United States of America · Full-time · Posted Sep 25, 2026
- Experience
- Typically 4–6 years building production ML/AI systems, including at least 1–2 years of hands-on generative AI and LLM-based application work.
- Education
- Master's or PhD degree in computer science, data science, mathematics, statistics, or a closely related quantitative discipline.
About the job
Build and operate production agentic AI workflows that help Marvis diagnose, troubleshoot, and resolve networking problems.
What you'll do
- Design and improve agentic workflows, tool-calling patterns, skills, and MCP server integrations.
- Develop retrieval, memory, and evaluation services; assess and improve agent quality.
- Deploy and operate containerized AI services, including CI/CD, observability, and performance monitoring.
What they ask for
- Production experience with agentic orchestration frameworks such as LangGraph, LangChain, Claude Agent SDK, or equivalent; hands-on MCP or equivalent tool-serving protocol experience.
- Strong statistical and machine learning foundation; experience building evaluation and testing frameworks for non-deterministic AI systems.
- Proficient Python development in production codebases; experience with Kubernetes, CI/CD, containers, backend APIs, and AI-system observability.
- Master's or PhD in computer science, data science, mathematics, statistics, or a closely related quantitative discipline.
Nice to have
- LangGraph is strongly preferred; experience with agent memory systems or sandboxed code execution environments.
- Networking domain knowledge; experience with AWS Bedrock, OpenSearch Serverless, or similar managed AI/ML services.
Pay
USD 155,500 - 315,000 (base salary)
Tools
LangGraph, LangChain, Claude Agent SDK, MCP (Model Context Protocol), Python, Kubernetes, Docker, FastAPI
AI focus
Develop agentic AI and generative AI capabilities for Marvis, including orchestration, reasoning, tool use, retrieval, memory, evaluation, and production operation.
Model-written summary · the original posting is authoritative