Applied AI Engineer

Sauna•San Francisco, CA

23h•Onsite

About The Position

As an Applied AI Engineer, you’ll be responsible for building, refining, and scaling the agent systems inside Sauna, from architecture to evals to deployment. We care about what works in production: fast response times, predictable behavior, traceability, and uptime. You’ll work across infra, frontend, and product to make sure the agents people build inside Wordware actually work. A few examples of what you might work on: Implement multi-step, tool-using agents that hit real APIs and handle retries, auth, timeouts, and edge cases. Design agent memory systems that persist relevant state across runs, e.g. memory migrations, context organization, and orchestration state. Create agents that proactively do work and send you reminders. Own and evolve our eval framework: both automated checks and human-in-the-loop scoring. Plus whatever else you see fit.

Requirements

Minimum 3+ years of engineering experience, including time shipping production software.
You've built and deployed agent-like systems: multi-step LLM pipelines, tool-using bots, scripted assistants, or similar.
Hands-on experience with: Agent orchestration and memory management (e.g. memory migrations, state organization).
Hands-on experience with: Tool use and orchestration (e.g. calling real APIs, using plugins, auth flows)
Hands-on experience with: Evaluation: success metrics, regression testing, and improving agent behavior over time
You write production-grade code and can work across systems without needing a spec.
You'd rather ship than polish forever.

Nice To Haves

Shipped agents that live in the wild, used by customers, not just internal demos.
Familiarity with LLM ops, tracing, observability, and failure handling.
You've been a founder or early engineer, and it shows in the bar you hold your own work to.

Responsibilities

Implement multi-step, tool-using agents that hit real APIs and handle retries, auth, timeouts, and edge cases.
Design agent memory systems that persist relevant state across runs, e.g. memory migrations, context organization, and orchestration state.
Create agents that proactively do work and send you reminders.
Own and evolve our eval framework: both automated checks and human-in-the-loop scoring.