PageFly — مولّد صفحات هبوط بالذكاء الاصطناعي

Maystro Delivery held an internal hackathon to showcase what agentic AI can do. In 24 hours, our team of four built PageFly: give it a product's name, price and photos, and it writes and builds a landing page to sell it. We placed second.

What I built

I designed the multi-agent workflow in LangGraph and built the agents. One analyses the product photos, one researches a marketing angle on the web, one writes the copy and one generates the page's HTML. The research step is conditional: when the seller already has a marketing angle, the graph skips it and goes straight to the copywriter. My teammates built the API, the import of products from a Shopify store, and the frontend.

Later

In October 2026 I came back to PageFly and prepared the backend for an open-source release. A check step now validates each generated page, and a repair agent fixes it once when the check fails. I added offline tests with a fake model, CI, security fixes in the Shopify import, and support for any OpenAI-compatible provider. The 24-hour version stays tagged in the repository.

What I took from it

PageFly was my first applied agentic-AI project; until then, I had only taken Google's 5-Day Gen AI Intensive course. It taught me to keep agents isolated, each with a clearly bounded job. Agentic AI had looked hard before; after PageFly, I was confident building it.