Principal AI/Agentic Engineer

What you'll do

• Design and operate production agentic systems including supervisor, planner, and multi-agent patterns with human oversight.
• Build durable, resumable workflows with persistent state, checkpoints, and human-in-the-loop controls.
• Engineer tool and function integrations with enterprise systems using schemas, authorization, and failure containment.
• Develop context-engineering strategies for retrieval, chunking, indexing, memory hierarchies, and structured outputs.
• Create evaluation systems using curated datasets, safety metrics, and LLM-as-judge programmatic evaluators.
• Instrument agent applications through registries and tracing services to monitor prompts, costs, and latency.
• Engineer for production reliability including streaming, concurrency, rate limits, and incident response.
• Hire, coach, and develop a team of agentic engineers while writing and reviewing production code in Python.

What we're looking for

• 10–15+ years of software engineering experience in distributed, API-driven, or workflow-intensive production systems and technical leadership.
• Hands-on experience designing, coding, testing, deploying, and operating agentic or LLM-based applications in production.
• Strong understanding of agent architecture, tool calling, durable state, multi-agent coordination, context engineering, retrieval, evaluation, observability, and safety controls.
• Proficiency in Python and at least one of TypeScript, Go, or Java, including APIs, distributed systems, CI/CD, and secure SDLC.
• Experience defining and using measurable quality, safety, reliability, latency, and cost indicators to improve systems through iterative releases.
• Bachelor's degree in a related field or equivalent professional experience.
• Demonstrated ability to lead engineers, make technical decisions under uncertainty, and communicate with technical and non-technical stakeholders.
• Experience with agent frameworks, durable workflow systems, or highly regulated environments like health tech (preferred).