Day 1-2: From idea to full-stack app When I saw the H0 Hackathon announced, I knew immediately what I wanted to build. As someone who researched AI-driven food intelligence for my MSc dissertation — building ingredient embeddings, graph neural networks, and novel recipe generation — I've always believed food AI could do more than just recommend popular recipes. It should work for everyone, including the 250 million people worldwide living with food allergies. Fable is an allergen-aware recipe discovery app powered by Epicure — the largest multilingual food embedding model ever built (4.1M recipes, 7 languages, 1,790 ingredients compressed into 2MB of vector space). What's working in 48 hours:
EU Big 14 allergen filtering with a hand-curated truth table for all 1,790 ingredients — catching edge cases like oat milk (safe for dairy allergy) and almond milk (unsafe for tree nut allergy) Three recipe modes: ingredient pairings, AI recipe generation, and Safe Foods Mode Safe Foods Mode — built specifically for MCAS and severely restricted diets. Users define exactly what they can eat and Fable generates recipes strictly within that list, with "liquid of choice" placeholders for users who can't even assume water is safe Kitchen management with area tracking (fridge/freezer/cupboard/pantry), use-by dates, bought dates with automatic shelf life calculation, quantities and subtypes Full AWS DynamoDB persistence — allergen profiles, saved recipes, kitchen contents all survive sessions Deployed live on Vercel
Tech stack: Next.js, TypeScript, Tailwind, AWS DynamoDB, Epicure Core embeddings, Anthropic Claude claude-sonnet-4-6 Live app: https://v0-allergen-recipe-app.vercel.app More features incoming this week.
Log in or sign up for Devpost to join the conversation.