Open role

AI Pipeline Engineer

New York City  ·  Part-time  ·  Remote or hybrid  ·  Monthly retainer

Riverbend Studio builds agentic systems that run real businesses: pipelines of language models and automation that do real operational work. We operate always-on pipelines that power a relationship-intelligence service and a recruiting business, both in production today. We are hiring one engineer to own the reliability of those pipelines as we grow.

The standard

Reliability is the feature we ship.

Most agentic pipelines are built to impress in a demo. Ours are built to run a business. A few principles define how we work, and they are exactly what we are hiring for:

• 01Plausible but wrong is a critical defect, not a tolerated trade-off. A system that fails loudly is safe. One that quietly returns the wrong answer is not. We build the checks that catch it before it ships.
• 02Determinism over luck. The same input produces the same output on every run. We pin the models, hold the temperature at zero, and remove the variance.
• 03Nothing starts over. A job that stops at four thousand of ten thousand resumes at four thousand and one. Every run is checkpointed, every result reproducible.
• 04Everything is traceable. For any output, we can name the input, the prompt, and the model that produced it. The system is auditable by design.

If that strikes you as common sense rather than overkill, we should talk.

The role

What you would work on

• Validation gates wired inside the model call loop, so invalid output aborts the batch on the spot with a clear, traceable error rather than slipping through.
• Resumable pipeline runs built on per-record checkpoint tables in Postgres, so a killed process picks up exactly where it left off.
• Pilot-before-wave enforcement on every batch operation, so no full run executes without a reviewed sample first.
• Provenance logging on every model call, recording input, prompt version, model, and output, so any result can be traced to its source.
• Hardened Playwright browser automation against third-party web portals: selector stability, form-fill verification, and the messy edge cases real websites throw at you.

What you bring

• Production Python: real packages, command-line tools, proper error handling, tests. Notebooks are where you prototype, not where you ship.
• Hands-on Anthropic or OpenAI API experience: structured output, tool use, temperature-zero discipline, pinning specific model versions for reproducibility.
• Fluency in PostgreSQL: schema design, migrations, state and checkpoint tables, cursor-based pagination. Supabase a plus.
• Browser automation in production with Playwright or similar: form fills, multi-page flows, real-world edge cases.
• You have shipped and operated real pipelines, and you can talk in specifics about how you made them trustworthy.

Bonus points

• Pydantic or similar for schema-validated structured output.
• Async Python for running pipeline stages in parallel.
• External API integration done right: rate-limit handling, backoff, verifying response shapes before trusting them.
• A background where data integrity is strict and silent errors are costly: fintech, healthtech, compliance, logistics, or similar.

The setup

You work directly with the founder, who has built and operated real businesses and cares about this craft as much as you will. Real ownership over the architecture, structured documentation and a working test suite already in place, and the freedom to make the system world-class rather than merely functional. Ten to twenty hours a week to start, as a monthly retainer, with room to grow as we scale. The role is based in New York City and can be remote or hybrid. This is an independent contractor engagement to start. We begin with a 90-day trial at full rate so we both find out if it is a fit.