Founding Engineer — AI-Native Mental Health Startup

Source: SeekLabs master req. Formatted for internal scoring. This is the scoring standard for this role.

• Company: Legion Health — YC-backed, AI-native telepsychiatry / mental health care startup (founded 2021, 11–50 employees, Healthcare Technology). Company HQ is Austin, TX; this role is San Francisco.
• Public title: Founding Engineer (YC-backed AI-Native Mental Health Care Startup) w/ 0.2%–0.8% meaningful early ownership
• Openings: 2
• Location: In-person at San Francisco HQ — no remote-only
• Employment: Full-time
• Salary band: $130,000 – $250,000 / year
• Equity: 0.20% – 0.80%
• Visa sponsorship: None available
• Relocation assistance: None (candidate may relocate at own cost; SF presence is required)
• Channel: SeekLabs
• Manatal Job ID: TBD
• Internal fee / contract: 6.25% of salary, 90-day guarantee

Legion has two separated engineering strengths today: a human-facing ship-fast builder (UI/UX, internal ops surfaces, fast iteration) and a distributed/async systems builder (background jobs, scheduled workflows, infra). The gap is someone who can build both the human-facing control plane and the underlying stateful automation — so they can safely run many agents/workflows at scale, see what's happening via monitoring/ops UI, and prevent bad states by construction (guardrails and invariants). Must ship end-to-end in their stack fast, with strong systems thinking and enough LLM product intuition to own and repair agent surfaces.

The role owns core product + agent systems end-to-end: workflows, agent behaviors, guardrails, integrations, the UX that makes it usable, and the reliability that makes it trusted.

• Backend: Node.js, TypeScript, Supabase (Postgres), AWS (ECS, Lambda, S3)
• Frontend: Next.js 15 (App Router), Tailwind, Vercel
• AI: OpenAI, Anthropic, tool-calling agents, embeddings + vector DBs, Langfuse-style observability
• Other: PHI security, audit trails, real-time schedulers, transcript ingestion

A candidate missing any of these scores below 75 unless an explicit alternative path applies.

1. Production shipping in core stack, last 3–5 years: shipped a real production product using TypeScript, Next.js, and Postgres (Supabase + Vercel adjacency are strong pluses). Not acceptable: "skills section says Next.js," side projects with 0–1 users, or frontend Next.js with the backend in Python/Django.
2. Real backend/systems competence: schema and relational data modeling, stateful workflows, transitions and invariants, APIs that reflect real constraints (not CRUD hand-waving).
3. State / invariants mindset: can turn messy workflows into explicit states, valid transitions, and invariants that cannot be violated (e.g. cannot double-book a provider slot; a cancelled appointment cannot trigger upcoming comms; agents cannot spam or duplicate).
4. Can build both surfaces: independently delivers human-facing UIs (ops dashboards, admin tools) and background orchestration (scheduled jobs, workflows, agent runs).
5. Heavy, habitual AI-native dev tool user: Cursor, Claude Code, Codex, Cline, etc., with real workflow opinions and concrete examples of how these tools accelerate shipping without slop.
6. Early-stage + SF + IC: excited about early-stage execution (not remote-only — SF or willing to relocate), and wants to execute as an IC (not transitioning into management).

• Previous early-stage startup or founding-engineer experience.
• Experience in regulated/high-stakes domains (healthcare, HIPAA, fintech).
• Experience building and monitoring production LLM features (tool-calling, RAG, agents).

• No SF presence (remote-only, cannot/will not be in-person at SF HQ).
• No work authorization without sponsorship (no employer sponsorship available for this role).
• No recent production shipping in Next.js + TypeScript + Postgres (last 3–5 years).

• Production LLM app engineering: ownership of prompting, context management, tool-calling patterns, evaluation/monitoring, debugging real production failure modes.
• 10k–100k user feature ownership: personally shipped and iterated features used by 10k–100k users recently.
• Performance fundamentals for chunky data: batching/pagination, relational query design, real-world performance debugging for dashboards and joins.
• Lightweight CI/testing that increases velocity: pragmatic Playwright smoke tests, GitHub Actions, focused tests for critical flows.
• 100+ GitHub stars (1,000+ is an extremely strong signal).
• Multiple hackathon / coding-competition wins as a meaningful contributor.

• Research/academic/healthcare-ML without production systems (papers/datasets/classifiers, Python-first ML pipelines, no product shipping or systems ownership).
• Frontend-only / UI-only with shallow API language and no data model or state thinking.
• No real proof of build: GitHub mostly coding challenges/school/take-homes; <100 contributions in the past year; one-commit repos; low-effort links; no demos/repos/LinkedIn.
• Stack misalignment in practice (no recent production Next.js + TS + Postgres).
• Planner/executive returning to IC; long time since writing production code; primarily managed orgs/roadmaps/outsourced teams.
• Long indie-hacker gap with minimal output (6–12 months ok with a strong story; 3–4 years with <$1k/mo traction is a strong negative).

• Short founder stint (~6 months) — must explain clearly.
• Low GitHub contributions (~100–500/year) — must explain with alternative proof-of-build (private repos, company policy, etc.).
• Enterprise background — depends on teams and how they shipped; fast-moving teams can be good, process-heavy orgs usually misaligned.
• Hackathon wins without verified real contribution — verify via commit history.

• Counts: production product shipped and maintained over time; evidence of iteration; real users and operational pressure; adoption or revenue is a strong plus.
• Does NOT count: one-off demos, school projects, take-homes, coding challenges, side projects with only the candidate as a user.

• High-yield archetypes: early-stage product engineers shipping TS/Next/Postgres with real users; technical founders/CTOs (4 engineers) still hands-on; engineers who built ops dashboards/control planes for stateful systems; engineers who shipped LLM product features in JS/TS stacks; visibly AI-native builders with real shipped output.
• Low-yield archetypes: healthcare ML researchers; frontend-only specialists; infra/SRE-only without product ownership; exec/VP/CTO profiles without recent hands-on shipping.

1. Share an example of shipping in high ambiguity and how you decided what to build first.
2. Describe any experience building or integrating LLM / AI-powered features into production.
3. Why do you want to work at Legion Health?

1. Systems/portfolio deep dive (45 min) — walk through 1–2 systems you've shipped; architecture, tradeoffs, failure modes.
2. Practical work trial (1.5 hrs) — short, realistic backend/LLM-systems exercise. No LeetCode, no puzzles.
3. Final onsite (1.5 hrs) — meet the team, pair on a real issue, talk through how you'd own a domain.

• Time-to-hire: 7–10 days. Start date: ASAP.

From the SeekLabs client channel:

• A prior candidate (Trevor P.) was liked by the client's Chief of Staff; the CEO felt there was real signal but not enough — candidate passed.
• SeekLabs admin (Aman Sood): the main gap was systems depth against their current bar. Direction going forward: keep sending candidates who are strongest on stateful workflow / backend ownership, with clear SF + comp alignment.

Implication for scoring: weight stateful-systems/backend ownership depth heavily; treat a thin systems story as a real downgrade even when product/UI is strong.