The "tastiest" result of useless fun and cross-collaboration
The Origins of MMMMMMai 🥦
MMMMMai is the brainchild of PinkPonyClub, a team formed at the Useless Fun AI Build-A-Thon, presented by Haystack, hosted by Cloudflare, and fed by AssemblyAI. Our team of solo software engineers and product designers met for the first time at the event, united by one mission: to create a fun and utterly useless AI-powered product. The result? MMMMMai, a recipe generator that crafts absurd recipes based on user input.
By the end, we were a tight-knit group of 3 software engineers and 2 product designers, collaborating to bring this concept to life. Our challenge? To use Cloudflare’s WorkersAI, deepset’s Haystack LLM, or AssemblyAI’s SpeechAI model to build a creative, technically complex, and – most importantly – useless product. Our product designers crafted a brilliant design, while our software engineers tackled the tech within the time constraint. One of our teammates pulled together a highly entertaining presentation for the judges and audience.
While we didn’t take home the top prizes, we’re proud of what we built and thrilled to share our absolutely useless creation with the world. So, welcome to MMMMMai – a project as much about collaboration and fun as it is about useless recipes.
If you ever attempt one of our culinary disasters, we’d love to hear about it! Drop us a note on our contact page or tag us on LinkedIn. Bon appétit...and good luck!
How We Built It 💻
We built our useless project using Cloudflare's WorkersAI to tap into LLaMA 3.0, which generated our recipes and Stable Diffusion XL to manifest these meals visually. Due to time constraints, we forked Cloudflare’s Hackathon Helper starter kit, which included a TypeScript-based API originally designed for an etymology app. We cusotmized this API to suit our use case and utilized prompt engineering to define a role system that processes content with the directive: “You generate bad recipes. The user will input ingredients, and you will provide the worst recipe ever.” The API processed user inputs and forwarded them to the LLaMa model, which generated the absurd recipe text. This text was then passed to the Stable Diffusion XL model to produce a corresponding image. BOth outputs were then displayed to the user, constituting the 1.0 version of the project.
MMMMMai 2.0 is a re-engineered and optimized iteration of the original hackathon project. Driven by the potential of the concept, I (Shane Jeon) took the initiative to rebuild the app using a modern frontend tech stack, including Next.js, TypeScript, React, and TailwindCSS, which aligns with my expertise. For an enhanced user experience, I integrated Lottie animations for recipe-loading visuals. The frontend architecture was redesigned to improve efficiency, and the app was deployed on Vercel to ensure high performance and scalability.
MMMMMai GitHub Repo
