An applied AI engineer — explicitly not an AI researcher or frontier-model builder (the CTO owns that). This person understands the moving landscape of AI/ML capabilities and techniques, is fluent with open-weight / open-source models Nickel can run on its sovereign on-prem boxes, and builds the agents that interact with those models, the clinical data, and the medical protocols injected into the system. Every agent lives in a harness with heavy guardrails and eval functions so it behaves as intended — and because this is human-in-the-loop medical AI on PHI, everything the agent consumes or concludes (summaries/interpretations surfaced to physicians — never autonomous decisions) must be fully traceable.
A big part of the job is the practical craft of agents: agent drift detection (understanding how/why an agent diverges from its original purpose, monitoring it, then recovering, decommissioning, or redeploying), and the disciplined use of agentic code generators to build agents with guardrails, evals, unit/integration tests at every step — plus what the harness must provide for orchestration, logging, traceability, and drift detection. Prior experience doing exactly this is great; if not, they'll consider someone who can clearly get there given time and self-directed research. Like the architect, this is an end-to-end IC. They should also understand the data foundation — especially how the knowledge graph is built and navigated (which slice makes an agent more effective and surfaces correct vs. incorrect conclusions) — though a data-retrieval API abstracts the storage layer beneath them.
What You'll Own
- Build agents that interact with open-weight/open-source models, clinical data, and injected medical protocols
- The harness around them: guardrails, eval functions, orchestration, logging, traceability
- Agent drift — monitoring, detection, recovery/decommission/redeploy
- Disciplined agent-building with agentic code generators + unit/integration tests at each step
- Effective knowledge-graph navigation to feed agents the right context
Requirements
- Applied AI/ML engineer (not a researcher) — strong grasp of the current model/technique landscape
- Fluent with open-weight / open-source models and running them locally
- Real agent-building craft: guardrails, evals, harnesses, orchestration, logging
- Agent drift detection + recovery/decommission/redeploy experience (strongly preferred)
- Disciplined use of agentic code generators with tests at every step
- Understands knowledge-graph construction/navigation and data retrieval
- PHI/regulated traceability awareness for agent data access
Who Will Thrive Here
- An applied-AI engineer who loves the practical craft of reliable, evaluable agents (not model research)
- Comfortable with open-source models, harnesses, drift, and eval discipline
- Cares about traceability and safety in a human-in-the-loop medical context
- Self-directed IC who'll research and build what's needed
No benefits listed yet.