Staff AI Engineer | Agentic Systems

We're building production-grade agentic systems that audit medical claims end-to-end — reading raw medical records, reasoning over coding and clinical guidelines, and producing defensible findings that hold up to clinical and regulatory review. Reaching human-expert accuracy on noisy, long-context documents is one of the hardest unsolved problems in applied AI, and the field is moving weekly.

We're hiring a Staff (L6) AI Engineer to set the technical direction for these systems. You'll own the architecture of our audit agents from the first design decision to production, decide how we measure whether they're right, and make the engineers around you faster and better. This is a hands-on role: you'll write much of the hardest code yourself while leading the design work several engineers build on.

• →Own the agent architecture. Decide the loop, the tools, the context strategy and the evaluation harness for our claim-audit agents. Choose between single-agent and multi-agent topologies, between LLM reasoning and deterministic post-passes, between retrieval and direct context loading — and defend each choice with data.
• →Set the evaluation strategy. Define, with our clinical and coding experts, what expert-level accuracy means in measurable terms. Build the eval infrastructure the team ships against, diagnose where systems fail, fix the root cause, and prove the fix moved the metric.
• →Engineer context at scale. Long, noisy medical records; citation-grounded findings; structured outputs and tool surfaces that models call correctly every time.
• →Lead through ambiguity. Turn open-ended problems into technical plans, lead projects that span several engineers and quarters, and make the calls on model choice, build versus buy, and the trade-offs between cost, latency and accuracy.
• →Run it in production. Own reliability, observability and cost control for LLM workloads that run on real claims every day.
• →Raise the bar. Mentor engineers, review designs and code, and set the team's standards for agent design, evals and engineering quality — including how we use Claude Code, Codex and similar tools to move faster than most teams can without them.
• →Become a domain expert. Healthcare claims, coding guidelines, and the medical record itself are unavoidable parts of the job. Strong engineers who lean into the domain become outsized contributors here.

• 8+ years of software or ML engineering experience, including 3+ years building applied ML / AI systems in production, with a Bachelor's in CS, Math, Engineering or a related quantitative field (Master's or PhD preferred).
• Staff-level technical leadership: you've set the architecture of a production system other engineers built on, and led work across a team.
• Hands-on generative AI experience: training or fine-tuning generative models, or building agentic systems that solve real business problems in production.
• Deep, hands-on understanding of agent loops — how a model decides to call a tool, how a tool result re-enters context, how loops terminate, where they fail.
• Hands-on experience with at least one major agent SDK — OpenAI Agents SDK, Anthropic SDK / claude-agent-sdk, LangGraph, or equivalent — and a well-argued view of the trade-offs.
• Working knowledge of how modern coding agents are built and how they engineer context, and fluency with Claude Code / Codex as a power user — including reading their source when needed to understand or extend behavior.
• Strong Python engineering that sets the standard for a team: clean abstractions, type discipline, async, tested code.
• Solid command of VS Code and git — branches, rebases, worktrees, conflict resolution, PR workflows.
• A bias toward measurement: you've built evals a team relies on, you don't ship without one, and you don't believe a number you can't reproduce.

• Prior work on long-context, citation-grounded systems where the model must point to evidence, not just answer.
• Production experience with caching, observability, and cost and latency control on LLM workloads at scale.
• Experience designing structured outputs (Pydantic / JSON Schema) and tool interfaces that LLMs reliably call correctly.
• Familiarity with reasoning models (o-series, Claude extended thinking, Gemini thinking) and a sense of when they earn their cost.
• Healthcare, legal, finance, or any other domain where "mostly right" is unacceptable.

• Document understanding (OCR, layout-aware models, table extraction).
• Vision-language models, multimodal retrieval.
• Published work or substantial open-source contributions on agents, evals or LLM systems.