Senior AI Engineer, Citizen Health

• You've built LLM-powered systems that went to production and had real users — not just demos and prototypes
• You understand that the hard part of agentic AI isn't getting a cool demo working, it's making it reliable at the 95th percentile
• You've dealt with the unglamorous parts: evaluation frameworks, prompt regression testing, cost optimization, and edge cases that only show up at scale
• You care about the craft of building AI systems, not just the hype cycle
• You've formed opinions about when to use agents vs
• Chains vs
• Simple prompts, and you can articulate the tradeoffs
• You're comfortable working in a startup where the stack is evolving fast and the best architecture for next quarter might not be the one you built last quarter
• 4+ years of software engineering experience, with 2+ years building LLM-powered applications in production
• Strong proficiency in Python and hands-on experience with agentic AI frameworks (LangChain, LangGraph, CrewAI, AutoGen, or equivalent)
• Experience designing and implementing RAG systems — including chunking strategies, embedding models, vector databases, retrieval evaluation, and hybrid search
• Track record building AI systems that take real-world actions (tool-use, function calling, browser automation, API orchestration), not just generate text
• Strong software engineering fundamentals — you write production-grade code with tests, logging, error handling, and clear abstractions. You're an engineer who uses AI, not a prompt writer who avoids code
• Experience with evaluation and testing of LLM systems — you've built evals, measured hallucination rates, and designed regression suites for non-deterministic outputs
• Comfort with ambiguity and fast iteration. You've worked at a startup or early-stage company and thrived
• Strong written and verbal communication — you can explain agentic architecture tradeoffs to a PM and debug a prompt chain with an engineer in the same afternoon
• Don't meet every qualification? No worries! We believe that passion, curiosity, and the right mindset are just as important as a checklist of skills. If you’re excited about what we’re building, we encourage you to apply

Desirable

• We're hiring AI Engineers to build the intelligence behind our AI advocate
• You'll design and implement the agentic LLM systems, RAG pipelines, tool-use frameworks, and evaluation infrastructure that make it effective, reliable, and safe
• This is not a research role — you'll ship production systems that real patients depend on daily
• You'll work closely with Product, Platform Engineering, and clinical advisors
• You'll report to the CTO
• Agentic orchestration — Design and implement the multi-agent architecture that powers our advocate's ability to reason, plan, and execute complex multi-step tasks (scheduling appointments, filing insurance appeals, researching treatment options) with appropriate guardrails and human-in-the-loop checkpoints
• RAG and medical knowledge systems — Build retrieval pipelines over patient medical records, condition-specific knowledge bases, and community-sourced data from 125+ rare disease communities. Ensure the advocate's responses are grounded in the patient's actual medical history, not generic information
• Tool-use and integrations — Build and maintain the frameworks that let our agent take real-world actions: browser automation, phone calls, form filling, document generation, and API integrations with healthcare systems. Each tool-use needs to be reliable, auditable, and scoped to the right permissions
• Conversation intelligence — Design the systems that make the advocate's WhatsApp conversations feel natural, contextual, and proactive. This includes long-term memory, conversation state management, personalization, and knowing when to surface information the patient didn't ask for but needs
• Evaluation and safety — Build evaluation frameworks for agentic outputs in a domain where "good enough" isn't good enough. Develop testing infrastructure for multi-step agent workflows, hallucination detection, medical accuracy validation, and regression testing across model updates
• Prompt engineering and model optimization — Design prompt architectures that balance capability, cost, and latency. Evaluate and integrate new models as the landscape evolves. Optimize inference costs — our unit economics depend on it