We're hiring an AI Engineer to build and harden the platform underneath Nava's LLM-based products. Building at the leading edge of AI products has taught us that what comes out of the box for LLM applications is not enough. Agents run long-running tasks, which changes how we deploy software and manage state. Context has to be shipped into zero-trust sandboxes on different platforms, then retrieved and refreshed on our own terms. Different models need to serve different jobs at a cost we can sustain as usage grows quickly. We built first versions of all of this ourselves, each for the product in front of us.
What makes this opportunity unique: you take those first versions and harden them into a platform that every Nava AI team builds on. This is protected work, not a ticket queue. It exists to reduce the unplanned platform work landing on the product engineers at the tip of the spear, and to create the magic that lets the rest of the company move faster with AI, with real employers, employees and Nava teams on the other end.
What you'll accomplish in the first year
- Harden the platform for reuse: Take the first-version capabilities built for one product (context management, agent state, sandboxed execution, model routing) and make each reusable across use cases beyond the one it was built for, so new AI features start from the platform instead of rebuilding it.
- Make long-running agents safe to run: Design the state management and deployment patterns so an in-flight agent task survives deploys, restarts and retries, and can be resumed, inspected and replayed.
- Own context in and out of the sandbox: Build one thoughtful mechanism for packaging context into zero-trust execution environments across platforms, and for ejecting and retrieving new context as an agent works.
- Bring cost per task down as we scale: Route the right model to the right job, orchestrate subagents so expensive models do only expensive work, and make token usage measurable and optimized as volume grows.
- Give product engineers their time back: Measurably reduce the unplanned platform work interrupting the AI product teams, so their sprints go to features rather than plumbing.
- Make the ecosystem maintainable: Document, test, trace and evaluate the shared patterns so a new engineer can adopt them in days and regressions are caught before users notice.
What you'll bring
We understand that your experience is more than just a list of requirements, so we encourage you to apply even if you don't meet all of the following bullet points.
- Hands-on ownership of the infrastructure under an LLM-based product that other engineers or users depended on: an agent runtime, context or state management, orchestration, or evaluation and tracing. Internal users and inherited systems count.
- Shipping LLM features to production and living with them: imperfect model behavior, latency, cost and the tradeoffs between them.
- Modern full-stack development in TypeScript and/or Python, integrating foundation-model APIs, tool calling and human review into an application.
- Practical evaluation and tracing habits: you check what the model did and learn from real use.
- Explaining your decisions to product, design and domain experts, and turning their problems into reusable capabilities.
No specific framework, model provider or minimum audience size is required.
Our stack
React, Node, TypeScript, Python, Postgres and AWS; multiple foundation-model providers; sandboxed execution environments for agents; Langfuse for tracing and evaluation. Coding agents are part of daily work here.
What you'll get
- Protected space to build the AI platform every Nava AI team runs on.
- Frontier problems (long-running agents, context management, model routing, token economics) with real users in healthcare, not demos.
- A small, senior team and a direct line to the leaders making product and platform decisions.
- A remote-first company with a mission to fix healthcare and the tooling to do the best work of your career.
How to apply
Apply below with your resume; we read every application. The process starts with a short intro call, followed by a few conversations with the team, including a hands-on coding session.
