Start from the decision you want automated.

Moderation rules are rarely written down well enough for a rule engine, and they're specific to you: what's fine on one platform gets someone banned on another. Nyckel learns your line from the calls your team already makes, and gives every decision a confidence score — so the obvious cases clear themselves and your reviewers only see the genuinely ambiguous ones.

Works on product photos, listings, documents, uploads, and messages — anything arriving in a steady stream that somebody currently sorts. The categories are yours, not a fixed taxonomy, so the output drops straight into the structure your system already uses.

Quality checks are a good fit because "acceptable" is easier to demonstrate than to define. You show examples of good and bad, and the model picks up the distinction — including the parts nobody wrote into the spec. Confidence scores matter here: borderline items go to a human instead of being waved through.

Routing is where the confidence score earns its keep. Set a threshold, auto-route everything above it, and put the rest in a queue. As your team works that queue, their decisions feed back in and the threshold's worth of traffic keeps growing.

Useful when the answer isn't one of a fixed set of categories but "which of these existing things is this most like." Matching a new support ticket to the one that already got solved, finding the product that fits a customer's description, spotting near-duplicates in a library.

Receipts, labels, forms, signs, screenshots — anywhere the information you need is trapped in a picture. Often the first step in a longer chain: read the text, then sort or route on what it says.