Agent Researcher

About Bunkerhill

Bunkerhill is a long-horizon agent for health system knowledge work. Health systems use our agent’s built-in capabilities for work that is performed consistently across customers, and use its tailored capabilities for work that is specific to them.

We are an in-person, in-office team in SF and NYC that exemplifies three core values: build for the patient, own it ship it, be bold together.

About the team

What makes health systems unique from other verticals adopting agents is the use of repetitive intelligence: the same closed-ended task is performed tens to hundreds of thousands of times per health system per year.

The team manages the agent lifecycle for closed-ended tasks that demand repetitive intelligence, from learning about a customer pain point to launching an agent that runs on every patient.

About the role

• Inform optimal agent design through experimentation
• Optimize agentic correctness, cost, and velocity
• Increase agent developer velocity
• Establish reliable connectors with third-party systems
• Own agent testing and evaluation
• Identify and advocate for platform improvements
• Identify opportunities that demand repetitive intelligence

Your background

• Strong technical foundation: comfortable writing code, collaborating in technical teams, and understanding core agent concepts such as delegation, memories, tool-calling, compaction, structured outputs, context engineering, and prompt design
• Strong scientific background: comfortable running experiments and with AI fundamentals such as dataset curation, training an AI system, designing evaluations and metrics, error analysis, and drawing sound conclusions from noisy results

Even better

• Experience building AI systems or agentic systems
• A CS, engineering, or math degree or equivalent technical experience
• Experience with healthcare data or systems, such as EHRs, clinical documentation, or HL7/FHIR
• Experience taking an AI system from prototype to production at scale
• Published research or open-source contributions in machine learning or AI agents