Floravert: Merging 350 Grower Stock Lists

Floravert buys from roughly 350 growers, and every one of them sends their stock list in their own format. Different columns, different naming, different units, sometimes a different language. Someone had to turn that pile into one sales list by hand, over and over.

You can’t write a parser per supplier when there are 350 of them and the formats drift anyway. You also can’t realistically ask 350 growers to start using your template.

So the app treats format detection as the actual problem. It reads whatever spreadsheet arrives, and AI works out the structure: which column is the product, which is the price, which is the stem length, which is stock. Once the columns are identified, ordinary deterministic code does the rest. It normalises the data, applies markup and rounding per customer, translates NL to FR for the French-speaking customers, and exports to Excel in one click.

That split is deliberate. The model decides what things are, and it never does the arithmetic. Letting it interpret the messy part while normal code handles the numbers means prices come out reproducible rather than plausible, which matters quite a lot when you’re sending them to customers.

Hours of manual work every week, replaced by one upload and a click.

This is the kind of internal tooling I build through Klappe Development.