Inspiration The inspiration for Legacy AI is deeply personal. It started with a photo of my grandmother's recipe for Ragi Dosa, written in her own hand in a faded, multi-page notebook. I realized that while I had the instructions, I was in danger of losing the most important part: the sound of her voice explaining it and the stories she told while cooking. I knew countless families had similar treasures—fragile notecards and spoken traditions—at risk of being lost forever. I was inspired to build a tool that could preserve not just the recipe, but the soul behind it.

What it does Legacy AI is a digital time capsule for family recipes. It saves culinary traditions by allowing users to either record a loved one narrating a recipe, or simply snap a photo of a handwritten card. Our AI pipeline then automatically transcribes the voice or text and intelligently structures it into a beautiful, functional digital format. From there, users can compile their recipes into a private family cookbook to print as a physical heirloom, or publish it to our Community Library to share and sell their family's heritage with the world.

How I built it I built this full-stack application using bolt.new as our primary development environment, which enabled incredibly rapid iteration. The entire backend is powered by Supabase, which handles our PostgreSQL database, user authentication, and storage for all audio and image files. The core AI magic uses a custom-engineered prompt for OpenAI's API to intelligently structure recipe data. The entire application is deployed live and continuously integrated using Netlify, making it instantly accessible to a global audience.

Challenges I ran into Our biggest challenge was teaching the AI to handle messy, real-world data. Initially, the AI struggled to read my grandmother's multi-page recipe, cutting off ingredients and misinterpreting her handwriting. This forced us to build a critical "review and edit" screen, creating a "human-in-the-loop" system where the user can easily correct the AI's initial transcription. I also faced a technical hurdle where our front-end code would crash because it expected a single recipe object, but the AI correctly returned an array of multiple recipes. This led us to build a more robust multi-recipe review screen, which ultimately made the app much smarter.

Accomplishments that I am proud of I am incredibly proud of successfully building a complete, end-to-end AI pipeline that takes raw, unstructured inputs like conversational audio and complex handwriting and transforms them into clean, structured data. Our proudest moment was successfully digitizing the multi-page recipe from my own grandmother's notebook, proving that Legacy AI can save the very memories it was designed to protect. I also designed an empathetic, "senior-first" user experience and built the scalable, full-stack foundation for a real, startup-ready application.

What I learned This project was a deep dive into the practical application of AI. Our biggest takeaway is the power of sophisticated prompt engineering—giving the AI a clear role, good examples, and strict rules is the key to getting reliable results. I also learned that the most powerful AI systems are those that assist and augment human intelligence, rather than replace it. Building Legacy AI taught me how to architect a full-stack system by seamlessly integrating services like Supabase and OpenAI, but most importantly, it reinforced our belief that technology is at its best when it serves a deeply human need.

What's next for Legacy AI Our vision is to make Legacy AI the definitive platform for preserving and sharing culinary heritage worldwide. Our immediate next steps are to launch native iOS and Android apps for a more seamless recording experience, build out a full suite of tools for our Community Library creators (including sales dashboards and reviews), and enhance the platform with new AI features like smart meal planning based on a family's unique collection of recipes.