Fox.ai

Inspiration

One of us co-runs DNT, a family apparel business selling on Shopee, Mercado Livre and TikTok Shop in Brazil. Launching a single product takes him about 8 hours, and when a listing sells, he still can’t tell whether it was the photo, the angle, the price or the channel. Another teammate runs a small business in Portugal and hits the same wall. Big brands solve this with constant experimentation: Amazon says A/B testing listings can lift sales by up to 20–25%. But that tool only exists on Amazon, only for registered brands with high traffic. The 3+ million sellers on Shopee Brazil get nothing, and in 59% of small Brazilian online businesses, one person runs the whole operation (Sebrae, 2026).

What it does

Fox.ai turns a product launch into a loop that learns. A seller gives a photo, the production cost and a margin range. Fox.ai then:

• pulls top competitor listings with Apify to see what's winning in that niche;
• generates a portfolio of genuinely different listings (angle, cover image, kit, price, acquisition channel), each tagged by the axis it tests;
• tracks sales and adjusts prices and promotions within the seller's margin guardrails;
• learns which axes win (e.g.\ ``on-model cover + gym angle beats product-only + casual'') and feeds that into the next launch.

Generating listings is cheap now. Knowing which one sells is the hard part, and that's what we built.

How we built it

Our architecture is centred around Factory, because we really wanted to work with droids, as they provide incredible flexibility. In our project, we use internal product/sales data as inputs coupled with analysis of competitor listings to create ads for the customer. We also track real time sales data, which we use to manage prices and promotions within a range of margin.

Challenges we ran into

Even though we had a lot of knowledge of our customer group, we took a long time to characterise problems, discuss and ideate. This was an intentional strategy, but it made it really difficult to pack all of the coding into less time. We also wanted fully automatic publishing, but marketplace API access requires partner approval that takes far longer than a weekend. We turned that into a design choice: a human approves every listing before it goes live, which also protects sellers from duplicate-listing penalties.

Accomplishments that we're proud of

Coming up with a real, documented problem that is also tractable. This was what allowed us to achieve so many things throughout the hackathon. From building a fully working prototype, to designing a bot that made the scrabble tournament final. We are also proud of how we dealt with disagreement in the team, as there was never any hostility and everyone was looking to learn through discussion.

What we learned

We learned that there is a whole world of tools to build exciting things. And that though they may sometimes seem intimidating, jumping straight in is the best strategy to get truly immersed and make the most of the experience.

What's next for FOX.ai

• Run Fox.ai on DNT's real launches for 30 days and measure the lift against our current process.
• Pilot it with 10 other fashion sellers on Shopee Brazil, where fashion is the platform's biggest category.
• Add Mercado Livre and TikTok Shop, then Portugal and Europe through our teammate's business.

Built With

• apify
• chatgpt
• claude
• droids
• factory
• figma
• next.js
• react
• typescript