Staff Forward Deployed Applied Scientist (Agentic AI)

Role overview Work directly with customers to turn complex business needs into production-ready agentic AI systems. The role spans applied research, architecture, deployment, and customer feedback, with a focus on multi-agent workflows that perform reliably on real workloads and help improve the underlying AI platform.

Responsibilities - Design and build multi-agent workflows, reasoning loops, and tool-using language model systems for customer problems. - Work with customer technical leaders to translate uncertain requirements into dependable AI solutions. - Prototype agent frameworks and harden them for production, including support for persistent state and memory. - Develop reusable evaluation methods and safety controls to measure accuracy, execution security, latency, and cost. - Feed deployment lessons and scaling issues back to product, engineering, and research teams. - Validate AI platform capabilities by deploying them on customer-like workloads and surfacing gaps or bottlenecks.

Requirements - At least eight years of hands-on experience in applied AI, machine learning, or data science. - Demonstrated experience building multi-agent orchestration, tool-use or function-calling pipelines, structured outputs, dynamic planning, and persistent memory. - Strong production Python skills; experience with Pydantic, FastAPI, asyncio, or PyTorch is relevant. - Strong customer-facing communication and the ability to explain technical trade-offs to engineering and executive audiences. - Ability to travel up to 30% for customer work, workshops, conferences, and collaboration.

Nice to have - A master’s or doctoral degree in computer science, AI, machine learning, or a related technical field. - Experience with deep learning frameworks, distributed training, transformer or state-space models, and agent frameworks. - Research publications or patents, and familiarity with current work in language and vision-language models or agent systems. - Experience tuning model workloads for different accelerator architectures.