Building the monitoring layer for autonomous agents in production.
Applied AI Engineer
We're a small, well-funded team solving a problem that's about to become urgent: when companies deploy fleets of AI agents, how do they know what those agents are actually doing? Our platform instruments agent behavior at scale—catching drift, detecting prompt injection, and surfacing failure modes that log files miss. You'll build the systems that make agent behavior legible. This means designing fast evaluation pipelines, shipping tools that run against live agent traces, and turning raw telemetry into signals that save customers from silent failures. The work sits at the intersection of LLM application engineering and observability infrastructure, and we're early enough that your fingerprints will be on the core product.
What they're looking for
- 1–4 years of hands-on experience building with LLMs—prompting, chaining, structured outputs, and evaluation
- Fluent in Python; comfortable shipping to production and reasoning about latency, throughput, and cost
- Intuition for how language models fail in the wild—you've debugged weird outputs and built guardrails
- Bias toward building internal tools that accelerate iteration: eval harnesses, replay infrastructure, annotation UIs
- Clear communicator who writes design docs and thrives in a small, high-trust team in San Francisco