Bangladeshi coder wins global AI hackathon

Rezaul Karim Arif, a Bangladeshi network engineer based in Melbourne, has built PixelFlow, an AI-powered web application that recently won the AI Showdown Hackathon, a global competition hosted by AI coding platform Lovable with support from Google, Anthropic, and OpenAI.

Originally from Chattogram, Rezaul moved to Australia in 2015 to study IT at La Trobe University. Currently working as a network engineer, he also runs an automation and AI development agency that works with both local and international clients. According to Rezaul, his background in freelance design helped shape the vision for PixelFlow, which he now plans to scale.

From ideation to execution, the project was completed during the AI Showdown Hackathon held on June 6–8, 2025. More than 239,000 apps were started, and 5,118 were submitted for judging. Rezaul's project stood out for both its technical design and real-world usefulness. He earned $20,000 in prize money, receiving $10,000 from Anthropic for best use of Claude 4 in the first round and another $10,000 for being the overall winner in the final. The event was fully sponsored for 48 hours, giving all participants free access to premium features and tools.

PixelFlow is an infinite mood-boarding app designed to help users quickly turn abstract ideas into visual layouts. Rezaul says that unlike other tools in the market like Milanote or Miro, PixelFlow is made to be simple and easy to use. "It's like a visual notepad," Rezaul explained, "You open it and start building without needing to learn the tool first." The app is aimed at designers, marketers, and creative professionals who want a quick and simple way to organise their ideas.

Read more

AI Hackathon 2025 spotlights Bangladesh’s role in global AI innovation

Built in only 24 hours, PixelFlow uses several modern technologies, including React 18, TypeScript, Supabase, and Fabric.js. Rezaul also added browser-based AI features such as background removal using Hugging Face's transformer model, and image generation using HiDream-I1 hosted on Google's Vertex AI. He used Anthropic's Claude 4.0 model throughout the development as his main tool for coding support.

In the project, Rezaul also used different custom AI agents. According to him, each agent worked like a team member in a software development project, playing roles such as tester, designer, or product manager. "I break big features into smaller tasks and guide the AI step-by-step," he said. With help from Lovable's AI platform and tools like Windsurf, he was able to finish the design and functions with very little manual work.

While many apps use AI, Rezaul believes PixelFlow stands out because of its smooth user experience. "Most moodboarding apps have too many features and are hard to use. PixelFlow is simple by design, with a good UI and UX."

Although Rezaul has received several investment offers after winning, he plans to continue building the project on his own first. "I prefer to build something strong before taking outside funding," he said. His goal is to add more useful features, turn PixelFlow into a full SaaS product, and explore real commercial uses for it in the growing AI creative tools market.

"I want more Bangladeshis to explore this field," he added. "It's not rocket science, just clever system prompting."