Staff Engineer - Agentic AI

Responsibilities:

• Architect and build production-grade agentic systems spanning LLMs, tools, backend services, and healthcare workflows.
• Own critical systems hands-on while setting architecture and engineering standards across teams.
• Define Arintra's approach to agent evaluation golden datasets, metrics, regression suites, release gates, and production monitoring.
• Establish patterns for deterministic, LLM-as-judge, human-review, and trajectory-level evaluation.
• Define standards for prompts, models, tools, and agent versioning.
• Drive architecture decisions across tool design, orchestration, context, memory, guardrails, and human-in-the-loop workflows.
• Build safe release mechanisms, including eval-gated CI, shadow deployments, canaries, and rollbacks.
• Drive reliability, latency, and cost improvements across production AI systems.
• Lead technical design and architecture reviews across teams.
• Mentor senior engineers and raise the overall engineering bar.
• Make and defend build vs. buy decisions across models, frameworks, and AI infrastructure.

Requirements:

• 9+ years of software engineering experience with deep ownership of production backend systems.
• Strong experience designing distributed systems and making architecture-level decisions.
• 1+ year building LLM agents in production with real users.
• Experience taking an agent from design through production, monitoring, and continuous improvement.
• Deep understanding of AI evaluation datasets, metrics, baselines, regression testing, and release criteria.
• Strong technical judgment around model selection, agent architecture, reliability, latency, and cost.
• Demonstrated technical influence beyond a single feature or team.
• Strong ability to work across engineering, product, and domain teams.
• Hands-on enough to build, review, and debug critical production systems.
• 9+ years of experience owning production backend systems with architecture-level decisions (any language or stack).
• 1+ year building LLM agents (LLM + tool calling + multi-step) in production with real users in the current or previous role.
• Personally owned at least one agent end-to-end (design, prompts and tools, evals, deployment, monitoring, post-launch fixes), and owned evals, release standards, or architecture across multiple agents or teams.
• Concrete eval evidence: a named metric, dataset size, eval tool, or release gate.