Al.ta Cucina

1.5 million followers. Zero revenue.

Al.ta Cucina was Italy's biggest cooking community on Instagram: 1.5M followers, thousands of recipes shared daily. All of it living on someone else's platform.

I joined as founding designer to help build their first app. Twelve months in, I was running product for a team of 8.

Recipe Challenges

The engagement problem wasn't content — it was motivation, so I turned posting into branded competitions creators could win.

Each challenge was templated — countdowns, recipe cards, reward cards, CTAs — so we could launch 35+ without redesigning each one.

Social voting drove the growth loop: a recipe needed votes to win, so creators shared their entries on Instagram and their followers downloaded the app to vote. Some of those voters became creators themselves, bringing their own followers next time. During challenge months, recipe submissions jumped 2–3x and active-creator engagement went 5–10x. Challenge-driven sharing drove a 10x spike in new-user installs.

Recipe Badges

Monetisation had to pay three sides at once — brands, creators, and users — without showing anyone an ad.

Brands wanted awareness. Creators wanted income. Users wanted good recipes.

I designed Recipe Badges, a pay-per-view sponsorship system. Brands pay per recipe view against a capped monthly budget, creators earn based on traffic, users see a native ingredient recommendation instead of an ad. The Al.ta Card dashboard shows creators their real-time earnings and sponsored recipe performance.

Top creators ended up earning up to €500/month, and the top-creator base doubled in our peak year.

One model, two features

One fine-tuned GPT-3, two very different jobs — a conversational "digital Nonna" and AI-assisted recipe authoring. In 2022, well before LLMs were everywhere.

I fine-tuned a GPT-3 base model on Al.ta Cucina's entire recipe library, then used the same model for both.

Nonna Maria ran on top with prompt chains the backend team and I built together: a "digital Nonna" who could answer recipe questions, suggest ingredient substitutions, walk users through techniques in conversational Italian. Onboarding even opened with a "don't believe everything she says" disclaimer — basically hallucination safety in 2022, before that was a recognised pattern.

The Recipe Importer was the bigger bet. With 40,000+ recipes in the library, authoring was slow and inconsistent — and that inconsistency quietly hurt search. The best recipes also already existed: food creators had years of them sitting on their Instagram pages.

So I used AI to automate the authoring itself. Steps used to go in one at a time; I made it so you could paste raw text and it split into clean, evenly-sized steps, consistent across the whole app. Ingredients became standardised and reusable across recipes — which let us add substitutions and even nutrition info — and titles could be optimised for SEO on request.

The full version — paste a social post, get a complete, searchable recipe — was built and ready before I left, though I didn't get to ship it. For a tiny Italian startup in 2022, it was an unusually early, and genuinely useful, application of AI.

Fixing how we shipped

My biggest non-design impact: rebuilding the delivery pipeline with the CTO so design and back-end ran in parallel — monthly drops became weekly.

Features were bouncing between stakeholders, design, and engineering with no clear handoff points. Back-end would start before design was finalised, and front-end would discover problems back-end had already built around.

So I set up an up-front alignment between me, as design lead, and the CTO, then put design and back-end on the same rail, working in parallel against the same spec. By the time front-end picked it up, both sides were already aligned. No more going back.

Engineers stopped rebuilding what design hadn't finalised.

I was the team's only designer for the first stretch, then hired and mentored our second — so design could scale past me, not just through me.