AI Engineer - Insurance Intelligence & Patient Verification

We're building the AI systems that automate the revenue cycle for healthcare providers — verifying coverage, generating clinical documentation, coding claims, and talking directly to insurance payers. We're hiring AI product engineers, not traditional enterprise software engineers and not ML research scientists. The people who do well here are compulsive experimenters. They've spent nights and weekends building things with LLMs, they have opinions about which models fail in which ways, and they'll happily run fifty variations of an approach before settling on one. They're comfortable being wrong quickly — and rigorous about proving when something actually got better. About the Role You'll own the agents that determine, before a claim is ever submitted, whether we've identified the right patient and the right insurance coverage. This is the front door of the revenue cycle: when it's wrong, everything downstream fails expensively. The work is a tight measure-diagnose-improve loop against real production outcomes. What Success Looks Like in the First 6–12 Months • Measurable lift in automatic insurance match rate and patient identity verification rate • Measurable reduction in cases routed to human review, without an increase in downstream denials • A trustworthy eval and shadow-testing setup so every change is provably an improvement What You'll Bring — Role-Specific • Entity resolution and fuzzy matching intuition — names, DOBs, member IDs, dirty demographic data, and all the ways real-world records disagree with each other • Comfort building hybrid systems: deterministic matching rules plus model-based reasoning for the ambiguous tail • Precision/recall thinking — you can articulate the business cost of a false match versus a missed match and tune to it • Experience working with unreliable third-party data sources: payer responses, timeouts, contradictions, silent failures What You'll Bring — Every AI Engineer at GHR • Prompt and context engineering as a discipline. Decomposition, structured outputs, tool/function calling, and controlling model behavior through context design rather than hope. You can explain why a prompt works, not just that it does. • An evaluation-first working style. You build golden datasets, offline eval harnesses, and regression suites before shipping. “The model seems better” is not something you'd say out loud. • Agent design and orchestration. Multi-step agent workflows, tool use, retries and fallbacks — and the judgment to know when a deterministic rule should replace a model call. • Working code-level fluency. TypeScript/Node and/or Python — enough to build, test, and ship in our monorepo. You don't need to be a platform architect. • Data reasoning. Comfortable in SQL and semi-structured formats (JSON, EDI-like data); you can dig into real production data to find out why something failed. • Judgment about accuracy vs. autonomy. You understand that in healthcare, a confidently wrong answer is worse than an abstention, and you design confidence thresholds and human-review paths accordingly. • Regulatory maturity. You're comfortable working inside HIPAA constraints (no PHI in logs, controlled model endpoints, auditability). Prior HIPAA exposure is a bonus; respect for the constraint is mandatory. Nice to Have • Healthcare eligibility and benefits verification experience (270/271 transactions) • Familiarity with payer directories, insurance discovery services, or clearinghouse workflows • Exposure to coordination of benefits, Medicare Advantage plan identification, or secondary payer logic • Background in claims denial analysis — knowing which front-end errors cause which rejections • Prior work in identity resolution, KYC, or master patient index systems Across all roles, we also value: • Experience with AWS Bedrock, or running models under enterprise/compliance constraints • Familiarity with eval frameworks and LLM-as-judge patterns — including their failure modes • Cost and latency awareness: model routing, caching, right-sizing models to tasks • Retrieval/RAG experience where it earns its place, and healthy skepticism about it where it doesn't • You've built and shipped something small and real with agents, publicly or internally What We Don't Require • Long tenure in legacy enterprise stacks (Java/.NET/mainframe, large monolith maintenance) • Model training, fine-tuning from scratch, or an ML/PhD research background • Deep DevOps or infrastructure specialization — we have that covered • A specific number of years of experience — a sharp 3-year engineer with a strong AI portfolio beats a 12-year engineer who has only used ChatGPT Compensation $200,000–$225,000 base salary, plus standard benefits. This band is uniform across our AI Engineering roles — these are peer-level positions with equivalent scope and autonomy. Where you land in the band reflects the depth and strength of your portfolio, not your years of experience or prior title. How to Stand Out Show us, don't tell us. A portfolio, a GitHub repo, or a work sample will tell us far more than a résumé. We especially want to hear about something you built with LLMs where your first approach didn't work — what you tried, how you measured it, and what you changed. Work Authorization Candidates must be legally authorized to work in the United States on an ongoing basis. At this time, we are not able to sponsor or take over sponsorship of an employment visa (e.g., H-1B, TN, O-1) for this role, either now or in the future. Equal Opportunity Employer We are an equal opportunity employer and consider all qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, age, disability, veteran status, or any other legally protected characteristic.