Riverbend Studio

We run agentic systems in production: pipelines of language models and automation doing real operational work, behind a relationship-intelligence service and a recruiting business that are live today.

The hard part, and the whole job, is reliability. Most agentic pipelines are built to impress in a demo. Ours are built to run a business, where a plausible-but-wrong answer is a critical defect, not a tolerated trade-off. We pin the models and hold temperature at zero, so the same input gives the same output. We checkpoint every run, so a job that dies at four thousand of ten thousand resumes at four thousand and one. We log provenance on every model call, so any output traces back to the exact input, prompt, and model that produced it.

You work directly with the founder, own real architecture, and have the room to make these systems world-class rather than merely functional. The test suite and the docs already exist. We want someone who treats "plausible but wrong" as a bug worth hunting.