Principal Engineer

Legion HealthAt Legion Health, we believe everyone deserves fast, affordable, and world-class health care, and we're developing the AI infrastructure to deliver it ourselves at world scale.Legion is building autonomous medical care (the AI doctor), starting with psychiatry. Our AI-native care-delivery platform currently automates 95% of the administrative work required for us to deliver direct patient care. We also recently became the first company ever to receive regulatory authorization to let AI prescribe psychiatric medications, allowing us to not only collapse health care's admin costs but also its clinical labor costs—shifting this industry's economics from humans to tokens.Our technical moat is hard to copy—we combine rich data, production AI, doctors in the loop, and end-to-end care operations to deliver measurably better care to a clinically complex patient population. This has helped us produce tens of thousands of total visits in just 2 years, all while holding ops costs flat and achieving industry-leading patient NPS and retention. While the last generation of healthcare startups made the existing system incrementally more efficient, Legion is rebuilding full-stack care delivery from first principles. Our vision? A 10X better patient experience that's higher quality, less expensive, and more scalable than ever before. And a 10X better clinician experience that frees our providers to focus on what matters most: caring for patients.Legion Health is backed by Y Combinator, leading venture capital firms, and founders from Function Health, Modern Health, Everly Health, Trusted Health, Clipboard Health, PatientPing, Sesame Care, Faire, EasyPost, and fuboTV.Join us as we build the future of health care: faster, higher-quality, and more affordable, powered by doctors and AI working together.The RoleHey, I'm Daniel G Wilson, co-founder and CTO at Legion. I studied applied math and CS at Princeton, led the Office Scripts and Power Automate integration at Microsoft, and did deep learning and RL research before that.I'm hiring a principal-level engineer to own the hardest systems in our care platform, whole. Not tickets, not a lane inside someone else's architecture: entire problem spaces, from "this is a mess" to "this is done and here's the doc."Here's the honest shape of the job. We run a regulated clinical product (we e-prescribe controlled substances) on an event-driven TypeScript platform, and we ship at unusual velocity with heavy agent leverage. The engineers who have thrived here share one behavior: I hand them a gnarly domain, they disappear into it, and they come back with it working, documented, and theirs. I don't need to check their work or meddle, because checking isn't unnecessary. That's the seat. The people who have struggled here waited for scoped tickets to tell them what to do—but that's what we have agents for.What You'll OwnReal domains, picked together in week one from this list, with written authority over them:The factory spine. Own the question of how we get maximum leverage in the era of agentic engineering. We ship 500–1,000 PRs a month with a lean team. Merge gates, migration safety, canary and rollback on a system where a bad deploy can touch patient care—this all works today. Your job will be to move us towards 10,000 while maintaining quality and rigor.The clinical work spine. Our event-sourced system of record for clinical work (orders, tasks, commands, receipts): the layer that lets AI agents and humans act on patient care with auditability and fail-closed checks. You'll own the guardrails that ensure agents can safely provide broader and broader components of care.Agent operations in production. LLM agents doing real clinical-adjacent work with tool use, checks, and receipts. You decide how autonomy expands and ratchets upwards as proof accumulates. Your agent harness will help prescribe psychiatric medication, legally, at the second company ever allowed to.The data and metrics estate. Warehouse, analytics platform, and the pipelines that tell us what actually happened.Architecture syncs with me. You'll have the standing I give someone who has seen more systems than I have: disagree with me, make the call in your domain, unpack it for the team when asked.What "A Fit" Looks Like, ConcretelyYou've owned architecture across multiple companies or many codebases: staff-plus IC, chief-architect, or platform-consulting background. You've seen a dozen ways teams ship and you have opinions with scars behind them.You ship daily at high volume with a fleet of AI agents. That's table stakes. In the interview we'll ask to see your actual setup and what you let it do unsupervised. "I use Copilot sometimes" is not this.When handed an ambiguous, ugly, load-bearing problem, your instinct is to take it whole: read everything, decide, build, write the doc, tell us what you decided. Not to schedule a meeting about scope.You communicate proactively in writing. The pattern we love: a Sunday text that says "took care of it, here's the PR and what I changed."You hold clinical-grade correctness without losing speed: auditability, HIPAA, fail-closed defaults, invariants over vibes.This is a senior seat. Typically that means 10+ years and multiple systems owned end to end, but we care about the scars, not the number. We have separate openings for earlier-career engineers; this is not that. We need someone who's seen some shit.This role is not for you if you want to manage instead of build, you need well-scoped tickets to be productive, or big-company pace feels normal to you.Our Tech StackTypeScript, Node.js, Postgres, Next.js on Vercel, AWS, production LLM agent orchestration and bespoke harnesses, tools, CLIs, skills, evals and observability, EHR and e-prescribing integrations, PHI-grade security and audit trails.Process (7 to 10 days)Process (7 to 10 days) Coffee chat (20 min). We discuss the hardest engineering systems you've owned, earned scar tissue, and whether there's mutual interest and fit at a high level.Systems deep dive (60 min). You walk me through one or two systems you owned for years: architecture, tradeoffs, what broke and what you changed.Take-home (async). You receive a simplified slice of our real repo and system surface, identify an area you'd like to shore up / build out / rearchitect, and execute. The deliverable will be a PR with an implementation (not a plan) and your agent session logs.[Alternate] Paid work trial (2 to 5 days, flexible around your current job). A real domain from our actual backlog, in the real repo, with your own agent tooling. This is the interview. It's how our best people got hired, and it's how you find out whether you want this.Interviews with founders and engineering. You'll meet the rest of the talented team you'll work with here at Legion.Decision within 48 hours. Written offer or a written no with the reason.Apply: or email me directly with a system you're proud of.Start date: ASAPRole Logistics, Compensation, and BenefitsJob type: Full-timeRole type: EngineeringUS Visa Sponsorship: YesSalary: $250,000 to $400,000 depending on experience, plus meaningful early-stage equity.Health Insurance: Medical, dental, and vision benefitsAdditional Perks: In-person retreats, meal stipends, etc.Time Off: Flexible, unlimited vacation policyWork hours: Flexible but demanding—because we are (literally) saving lives.Work Setup: In-person @ San Francisco HQ; partial remote considered for exceptional candidates (our track record includes both).Impact: Work on something that truly matters to millions of patients and families.Autonomy and Learning: Ownership