Finta uses AI across its accounting platform for startups, with one rule: accuracy over automation. Every day, Finta uses Respan to test prompts across models, manage golden datasets, and run evals before shipping. See how Andy and his team build reliable AI accounting ↓

@andywang Thanks to Andy and the Finta team for doing this!

@finta @andywang Going from not knowing what happened to running tests daily is a big shift in how a team operates.

@finta @andywang Watching someone walk through their own workflow tells you more than any feature list!

@finta @andywang Prompt versions, evaluators, and thresholds all reading from the same production data is what makes this loop possible.

@finta @andywang "it was a struggle to know what was happening" is close to word for word what we hear on most first calls.

@finta @andywang Adding new evaluators as you find new failure modes is how this should work. The test set grows with the product.

@Andydy42 @finta @andywang killer product!

@finta @andywang Hearing how a team actually uses the product day to day is different from describing features. Good one to watch.

@finta @andywang The best sign a product works is when someone uses it for hours a day without being asked to.

@finta @andywang The evals-before-shipping bit is the actual product moat here. Reliable AI accounting won’t come from the flashiest demo, it’ll come from fewer silent mistakes.

@finta @andywang Running tests against a threshold before shipping is the part most teams skip. Good to see it as a daily habit here.

@finta @andywang Prompt versions and eval scores tied together is what makes a threshold mean anything. Otherwise you're just tracking diffs.

@finta @andywang The golden dataset + eval loop is the part most AI demos skip. Shipping fast is cute, shipping without regression roulette is the actual product.