We build automations under your name.

How it works

Scope

New builds, and the unglamorous fixes on existing ones.

Price

Fixed per build, agreed before we start. No hourly billing.

Timeline

Typically 3–5 days for a standard build.

Included

Tests, error handling, and a handover document — not extras.

Credentials

Sandbox by default. Dry run before any destructive write.

Visibility

None. The work ships as yours.

First one small, so you can judge the quality with nothing at stake.

What we're good at

Workflows that run green and still lose data. A hard crash kills the process before the error handler can fire; a renamed field writes an empty value and reports success; a paginated API returns part of the set and no error at all. These don't show up as failures — they show up as a client asking why last month's numbers are wrong.

That's the class of problem our background is built on: 10+ years of backend engineering, 7 of them on high-load production systems, and dozens of third-party integrations across REST, SOAP and JSON-RPC.

Public work

• automation-diagnostics

  Single-file diagnostics for automation failures that run green — token cost inside loops, silent workflow deactivation, and a resilient poller for APIs that paginate badly and time out. Each tool ships with a mock API and tests.

• dirty-integration

  A reinforcement-learning environment where an agent has to pull a complete dataset out of an API that misbehaves the way real ones do: repeating cursors, drifting types, quotas without Retry-After. 43 tests, three reference solvers, and a difficulty curve measured over 60 seeds per cell. Published on the Prime Intellect Environments Hub.