Founding Engineer—Build AI-Native Ops for Mental Health (YC S21, $1M+ ARR) at Legion Health | Y Combinator's Work at a Startup
# **Founding Engineer—Build AI-Native Ops for Mental Health (YC S21, $1M+ ARR)**
---
### **🧠 Build AI-native care infrastructure. Ship for real patients. Operate at speed.**
---
### **ℹ️ TL;DR**
We're building the **AI-native operations layer for psychiatric care**. Not diagnostics, but what happens **_outside_** the visit—the real operational backend: scheduling, documentation, billing, intake, risk detection, and more.
If you want to build core infrastructure with real AI, **own** systems end-to-end, and work directly with a deeply technical founder still up to his neck in the code—read on.
---
### **🗺️ Context**
Hey—I'm Daniel, co-founder and CTO of Legion Health (YC S21).
Mental health care is **operationally broken**—patients ghosted, clinicians buried in forms, payers chasing missing notes.
The industry is flooded with AI startups trying to automate away diagnosis—and even providers as a whole. **But diagnostics aren’t the bottleneck. Operations are.**
So, we’re solving the _real_ problem:
> "What if operations worked?”
We’re building a real-time, AI-powered backend for mental health clinics—**LLM agents + structured systems** that coordinate human care like it’s software. Think:
* Automated intake that personalizes itself mid-call
* AI copilots that document visits while verifying insurance
* Risk detectors trained on full transcripts and events
* Schedulers that close the loop without human friction
And we’re already live. Our agent infrastructure supports over **2,000+ patients** — with **only one support human**.
**Unlike most AI startups, we are our own customer**. We operate our own large psychiatric practice. The systems you build directly impact our clinicians and patients _today_. No theoretical B2B pitches, begging for pilots, and months of stakeholder alignment and bureaucracy—just real-world, high-stakes operational challenges.
⭐ If you've felt like AI is being wasted on toy tools or B2B busywork—**come work on something real**.
---
### **🏆 What we’ve already built**
* $1M+ ARR, growing fast—**post-PMF, pre-scale**
* $6M raised, **$3M in the bank**
* A live agentic co-pilot (think “Cursor for care ops”) that actually reduces admin overhead
* A functioning psychiatric practice with real clinicians, patients, and claims
* A high-leverage AI architecture in motion—LLM tool use, RAG, event-driven infra
* A clear path towards for an event-driven, simulation-capable architecture that self-improves
* A small, intense team (\~11 people) shipping daily—no dead weight
This is the moment where **foundational hires make the company**. You're not joining an idea-stage pipe dream or a late-stage dinosaur. You're helping us define the future.
---
### **👷🏻♂️ What you’ll build and own**
This isn't a feature factory. We move **_fast_**, shipping multiple meaningful features straight to real patients and providers every single day.
You’ll **own entire domains** end-to-end—architecture, implementation, iteration—not just JIRA tickets. This is the frontier of applied AI in healthcare.
* **Core Event-Driven Backend:** Architect and scale our Node.js / Next.js / TypeScript / Supabase (Postgres) / AWS stack. Design schemas, event flows, and APIs that power real-time, resilient psychiatric care ops.
* **LLM Agent Infrastructure:** Build _actual agent loops_—tool use, memory, retry logic, context updates, feedback mechanisms. Not a demo. A co-worker.
* **Human + AI Ops UX:** Shape real-time interfaces where human teammates and agents co-work, co-learn, and co-adapt. Agents learn by using the same UI our humans do.
* **World-State Simulation:** Define the canonical state of a patient’s journey. Power alerting, planning, and agent behavior with a simulation of psychiatric care at scale.
* **Data & Compliance:** Engineer secure, HIPAA-compliant pipelines for transcripts, events, and EHR data—structured for both operations _and_ AI training.
* **System Design & Strategy:** You’ll work directly with Danny (CTO) to debate architecture, invent new primitives, and define the foundation for AI-native mental health systems.
You’ll help answer questions at the frontier of AI, like:
> _What does reliable agent infrastructure look like in production?_
>
> _What’s the role of structured data in a world with LLMs?_
>
> _How do we make agents auditable, evolvable, and fast?_
---
### **🫵🏻 This is for you IF**
* You've built and owned complex systems from 0 to 1—not just features but foundations
* You think in systems, state machines, and event flows—not just endpoints.
* You're LLM-fluent or are a strong systems engineer eager to become fluent _fast_.
* You default to velocity and appreciate clean architecture. Move fast _and_ build resiliently.
* You're allergic to bureaucracy, performative work, and slow decisions.
* You hold an incredibly high bar for yourself and expect it from others. Mediocrity is painful.
* You want direct impact, technical depth, and to solve problems that have never been solved before.
* You believe AI today is the worst AI will ever be—and want to be a part of building the future.
We're not looking for warm bodies. We’re looking for **founding technologists** who can _think in systems_ and ship fast.
---
### **🙅🏻♀️ This is NOT for you IF**
* You need extensive structure, mentorship programs, or a predefined career ladder.
* You prefer working on a single, well-defined component.
* You view LLMs as just another API call.
* You're uncomfortable with ambiguity or rapid iteration.
* You can't handle direct, honest feedback or thrive in a high-candor environment.
---
### **⚙️ Our Stack**
**Backend:**
Node.js, TypeScript, Supabase (Postgres), AWS (ECS, Lambda, S3)
**Frontend:**
Next.js 15, Tailwind, Vercel
**AI Infra:**
OpenAI, Anthropic, agentic loops and workflows, observability and eval (e.g. Langfuse), embeddings and vector DBs, tool-calling, model context protocol (MCPs), etc.
**Other:**
PHI security, audit pipelines, real-time schedulers, transcript parsing
---
### **🚀 Why Legion? Why Now?**
This is the moment when the infrastructure is still malleable. When the next 20 years of mental health care can still be shaped by a few engineers with systems taste, speed, and conviction.
You won’t be joining as employee #73. You’ll be founding the engineering culture. You’ll have a direct line to me. You’ll shape core systems _and_ help decide what we build next.
If you’ve ever said, “I wish I could’ve been there when \[insert legendary product\] was getting built,” this is that moment.
**If this resonates, I want to work with you**. Let’s build the founding systems of AI-native mental healthcare—and make something people didn’t think was possible.
Y Combinator's Work at a Startup 
