Senior Software Engineer

We're hiring a senior full-stack engineer who has made AI a core part of how they build software, not a novelty. This is not a machine learning research role. You won't train models or tune GPUs. You will ship product in a modern TypeScript/Next.js stack, and you'll do it with AI woven into both the product and your own workflow.

• Own features end to end across our Next.js/React/TypeScript app, Prisma/Postgres backend, and AWS infrastructure.
• Extend our AI coach and build new LLM-powered member experiences: retrieval over benefits data, structured extraction, guided flows, and agentic tool use against our own APIs.
• Design evaluation and observability for AI features so we know when they get worse, not just when they crash.
• Push our AI-assisted development practice forward: agent instructions, reusable skills, hooks, MCP integrations, and review practices that keep quality high as velocity climbs.
• Handle sensitive health data responsibly. PHI boundaries, redaction, audit trails, and vendor data agreements are part of the design, not an afterthought.
• Review code, whether written by a human or an agent, with a critical eye. Mentor teammates and raise the bar on testing and reliability.

• Healthcare, benefits, or other regulated-data experience (HIPAA, SOC 2).
• Building MCP servers or internal agent tooling for a team.
• Experience with Clerk, Stripe, or Sentry.
• Contributions to AI developer tooling, open source or internal.

Marvin · Denver/Los Angeles – Hybrid 2-day in office · Full-time · $160k-190k

• You build with AI. You've moved well past autocomplete. You direct agentic coding tools to do real, multi-step work, you know exactly where they break, and you own every line they produce.
• You build AI into the product. You've shipped LLM-backed features to real users and lived with them in production: prompts, tool calls, evals, tracing, cost, latency, and failure modes.

• 5+ years building production web applications, with deep TypeScript, React, and Node experience. Next.js is a strong plus.
• Comfort across the stack: relational data modeling (we use Prisma and Postgres), auth flows, caching, background jobs, and cloud deployment (AWS, Docker, Terraform or similar).
• A real testing habit at every layer: unit, component, and end-to-end. We use Vitest, Storybook, Cypress, and Playwright.
• Strong product judgment and the ability to work directly with design, clinical, and partner-facing teammates.

• Daily, expert use of agentic coding tools such as Claude Code, Cursor, or Codex. You write project instructions and reusable skills for your agents, hand them multi-step tasks, and review the output like a senior reviewer, not a rubber stamp.
• Hands-on experience shipping LLM features: prompt design and versioning, structured outputs, tool/function calling, streaming UIs, retrieval-augmented generation, and multi-turn conversation state.
• You've run evals and traced production LLM traffic (LangSmith, Braintrust, Langfuse, or homegrown), and you can explain what changed the last time a model version bumped underneath you.
• You understand the cost, latency, and quality triangle and have made real trade-offs: model selection, caching, prompt compression, and fallback strategies.
• Clear-eyed about limits. You know when an LLM is the wrong tool, how to constrain one when it's the right tool, and how to keep a hallucination from ever reaching a member.

160-190k base salary · Equity · Health, dental, and vision · [401(k)] · Unlimited PTO · Hybrid 2 days/wk in office (LA or Denver)

About Marvin

Marvin helps people navigate their mental health and wellbeing. We partner with employers and health plans to give their members one place to understand their coverage, find care, and get support, including an AI coach built into the product. We're a small team shipping a production healthcare application, which means every engineer touches things that matter and owns them end to end.

How we work

AI is part of our engineering culture, not a side experiment. Our repository ships with agent configuration, we maintain shared skills for code review and security checks, and we expect engineers to spend more of their time on judgment and less on typing. We also expect the opposite discipline: no unreviewed AI-generated code lands, and "the agent wrote it" is never an explanation for a bug.