Skip to main content
EigenPal automates repetitive work with AI: the tasks a person or a one-off script would otherwise do over and over. Document processing is the most common kind, and the examples here use it, but the same applies to any repeated task you can describe as steps. It exists to fix two problems that show up the moment that automation meets production.

​
The same work gets rebuilt and redone

Repetitive work tends to get re-implemented by every team that needs it, as brittle scripts or manual steps, each slightly different. Document processing is the clearest case: the same file is OCR’d and extracted again and again by different applications, each paying the cost and getting slightly different output. EigenPal makes the work a reusable artifact. You build it once as a workflow, version it, and reuse it everywhere. For documents, the result of processing a file is itself cached and reused, so the expensive steps happen once. See Everything is an artifact.

​
You can’t assume the output is correct

An automation that works on one input is not done. Models drift, edge cases appear, a vendor changes behavior, and a workflow that silently starts producing the wrong answer breaks every process downstream of it. Spot-checks and “looks right to me” do not scale, and they give you no way to prove quality to the teams that depend on the result. EigenPal treats quality as something you measure, monitor, and govern, not something you hope holds. You attach a dataset and evaluators that define what correct means for the task, tune until it passes a threshold, and keep those same checks running for the life of the workflow so a change can’t quietly regress it. See Evaluations.

​
The bet

Put together, those two ideas are the bet EigenPal makes: AI automation should be built like software, with the same discipline around correctness. Version it, test it, review it, deploy it, measure it, and improve it from real usage. That is what the rest of these docs describe. Start with How it works.