Workflow evals

What it does

Resources for evaluating workflows built with TypeSafe.

How you can use it

Try turning one of your project's written policies into small decisions. For expense claims, ask whether the receipt is readable and what kind of expense it shows. Let a developer write the rules for amounts and approvals, while AI answers the questions that need a judgment about the text.

Start by marking each sentence in your policy as either a rule or a question. Then compare the results with answers you trust. This site's comparison uses answers from two other AI models as its answer key. That helps compare systems, but does not prove the answers are right.