SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Nava is a fast-growing benefits brokerage fusing deep healthcare expertise with cutting-edge AI technology. The company is on a mission to fix healthcare by delivering modern, transparent, and affordable benefits experiences to employers and employees. Nava has built proprietary AI-powered products (like HQ, an AI-driven benefits platform) and is now seeking an AI Engineer to build and harden the platform infrastructure underlying these LLM-based products.
In this role, you will take first-version capabilities built for specific use cases and transform them into a reusable, hardened platform that all AI teams at Nava build upon. Your work will focus on reducing unplanned platform work for product engineers and enabling the company to move faster with AI.
Key responsibilities include:
- Hardening platform capabilities (context management, agent state, sandboxed execution, model routing) for reuse across multiple AI products and use cases
- Designing state management and deployment patterns so long-running agent tasks survive deploys, restarts, and retries, and can be resumed, inspected, and replayed
- Building mechanisms for packaging context into zero-trust execution environments and retrieving/refreshing context as agents work
- Optimizing cost per task by routing the right model to the right job and orchestrating subagents efficiently
- Reducing unplanned platform work interrupting AI product teams so they can focus on features
- Making the ecosystem maintainable through documentation, testing, tracing, and evaluation
You will work on frontier problems (long-running agents, context management, model routing, token economics) with real users in healthcare, not demos. The team uses React, Node, TypeScript, Python, Postgres, AWS, multiple foundation-model providers, sandboxed execution environments, and Langfuse for tracing and evaluation.
Requirements:
- Hands-on ownership of infrastructure under an LLM-based product that other engineers or users depended on (agent runtime, context/state management, orchestration, evaluation, or tracing). Internal users and inherited systems count.
- Shipping LLM features to production and living with them: understanding imperfect model behavior, latency, cost, and tradeoffs
- Modern full-stack development in TypeScript and/or Python, integrating foundation-model APIs, tool calling, and human review into applications
- Practical evaluation and tracing habits: checking what the model did and learning from real use
- Ability to explain decisions to product, design, and domain experts, and turn their problems into reusable capabilities
No specific framework, model provider, or minimum audience size required. The company assesses comparable work and reasoning.
