Growth Operator
A brand's ad account already tells you what is wrong, but the evidence is scattered across public sources and nobody reads it before making creative.
Architecture
Overview
Growth Operator reads what a Shopify brand is already doing in paid media, using only public data, and turns that into a diagnosis before it produces anything. It names the state of the account, shows every number with a link back to the source, and lets that state decide the creative direction. Built for the Shopify hackathon, it is the productized version of the work I do by hand for e-commerce clients.
How it works
The chain runs in six stages and each one writes its result, so a run is replayable from any point. Collection pulls the brand's live ads from the Meta Ad Library, its catalog from Shopify and its positioning from the site. Diagnosis classifies the account into one of seven states, volume without a winner, catalog only, retail push, just launched, and so on, with a confidence level that drops when a source degrades.
The architectural decision that matters is in the third stage. It chooses nothing on its own, it reads the verdict of the diagnosis. If the current ads are not performing, the system does not replicate that ad library, it goes looking for angles elsewhere. That rule is the actual product.
Production then extracts the brand's art direction from its own running ads through vision, measures the real palette, and generates creatives anchored on it. Each one ships with the line explaining why that concept, given the diagnosis.
Why I built it
Most creative work in e-commerce starts from taste and ends in a report nobody acts on. I wanted the opposite. Every number in the output carries a proof link the owner can open in the Meta Ad Library and check without me, and the run ends with an artifact they can deploy on their own. Six real brands are recorded as fixtures covering five of the seven states, so the whole chain replays offline against genuine data rather than invented examples.
Try it
The demo runs on recorded fixtures from real brands, so nothing depends on live API keys. Pick a brand, read the diagnosis, follow one of the proof links back to Meta and check the number yourself.