Preplo

Inspiration

We've all been there. A friend sends a cooking video at 11 PM — "you HAVE to make this." You save it. You forget about it. Weeks later, you find it buried in your camera roll alongside 14 blurry screenshots of ingredient lists you'll never decode.

Over 70% of Gen Z and Millennials discover recipes through short-form video — YouTube, TikTok, Instagram Reels. But the gap between watching a recipe and cooking it is massive. You have to pause, rewind, screenshot, guess quantities, and somehow remember which step needed "a pinch of that thing he added at 4:37."

We built Preplo because we were tired of saving cooking videos we never actually cooked.

What it does

Preplo turns any cooking video into a real, actionable recipe in seconds.

Paste a link. Get the recipe. Just cook.

Instant Recipe Extraction — Drop a YouTube, TikTok, or Instagram Reel URL. Our AI model analyzes the video and extracts a fully structured recipe: ingredients with exact quantities, step-by-step instructions with video timestamps, estimated costs, and nutritional info.
Smart Recipe Alterations — Tap "Make it Vegan," "Low-Carb," or "Extra Spicy" and the AI regenerates the entire recipe with updated ingredients, adjusted cooking times, recalculated costs, and modified nutrition. Each recipe comes with contextual alteration suggestions tailored to its cuisine and ingredients.
Guided Cook Mode — A full-screen, hands-free cooking experience. Large text for kitchen visibility, text-to-speech that reads each step aloud, embedded video with timestamp sync, and screen-awake mode. No more touching your phone with flour-covered hands.
Smart Shopping Lists — Auto-generated from any recipe, grouped by category (produce, dairy, meat, pantry), with price estimates. Merge multiple recipes into one trip-ready list for the week.
Cooking Streaks — A Duolingo-inspired engagement system that turns cooking into a habit. Weekly streaks, tier progression from Beginner Cook to Chef Legend, streak freezes, and milestone celebrations.

How we built it

Preplo is a full-stack system with three major layers: a cloud backend, an AI extraction engine, and a cross-platform mobile app.

The Backend is a modular API server built on NestJS with a PostgreSQL database and Redis caching layer. It's organized into isolated service modules — video processing, recipe management, AI extraction, engagement tracking, media storage, authentication, and subscription management — each with its own controllers, services, and data models. A custom JWT-based auth system handles user sessions, and a premium gating layer controls feature access across the entire stack.

The AI Extraction Engine is the heart of Preplo. When a user pastes a video URL, the system identifies the platform, processes the video through a platform-specific analysis pipeline, and sends the extracted content to our AI model. The model analyzes the video and returns a fully structured recipe — parsed ingredients with quantities and categories, ordered instructions with timestamps, cost estimates, difficulty rating, cuisine classification, nutritional breakdown, and suggested recipe alterations. We built a multi-layer validation system on top of the AI output to ensure consistent, accurate results regardless of the source video's language, format, or platform. The entire pipeline supports automatic translation, unit standardization, and intelligent failure detection when the video isn't actually a recipe.

Recipe Alterations run through a separate AI pipeline that takes the original extracted recipe, understands its full context, and generates a completely new version based on the requested modification — not a simple find-and-replace, but a full re-interpretation of the recipe with recalculated ingredients, updated instructions, adjusted costs, and modified nutrition.

The Caching Architecture is designed for scale and cost efficiency. We built a multi-layer caching system: a global recipe cache that stores every AI extraction so the same video never gets processed twice, an alteration cache that stores every recipe variation, a media pipeline that optimizes and serves thumbnails through a CDN, and a transcript cache for faster reprocessing. The result is that popular videos are served instantly at zero marginal cost, and the economics get better — not worse — as more users join.

The Mobile App is built with Expo and React Native for cross-platform delivery. It uses file-based routing, a dual state management architecture (client-side stores for local state, server-synchronized queries for remote data), and a custom animation layer for smooth 60fps interactions throughout. The cook mode is a standalone experience with text-to-speech, video synchronization, gesture navigation, and session tracking.

The Engagement System is a custom-built streak engine inspired by Duolingo. It tracks weekly cooking activity across timezones, manages streak freezes with automated monthly resets, and drives users through a five-tier progression system — all built from scratch without third-party gamification SDKs.

Challenges we ran into

Making AI output reliable and consistent. Early versions of our extraction pipeline produced wildly inconsistent results — missing ingredients, hallucinated quantities, instructions in the wrong language. We had to build an extensive prompt engineering framework with strict output rules, layered validation, and failure detection to get results that users could actually trust and cook from.

Supporting three video platforms with one seamless experience. YouTube, TikTok, and Instagram each work completely differently under the hood. Short URLs need resolution, some platforms block cloud server IPs, and the content format varies wildly between them. We built platform-specific processing pipelines that all converge into a single, unified recipe format — so the user just pastes a link and it works.

Making recipe alterations feel instant. Generating a full altered recipe through AI takes time. We designed a shared alteration cache so that once any user generates the "vegan version" of a popular video, every future user gets it instantly. The challenge was designing the caching strategy to maximize hit rates while keeping recipes fresh.

Building a streak system that motivates instead of punishes. Our first implementation was too strict — miss one week and you lose everything. We iterated to a forgiving model with streak freezes, weekly cadence instead of daily, and celebratory milestones that make users feel rewarded rather than anxious.

Accomplishments that we're proud of

We solved a problem we actually have. This wasn't a hypothetical project — we built Preplo because we personally needed it. Every feature was born from a real frustration we had in the kitchen, and that kept us honest throughout.
Three platforms, one paste. YouTube, TikTok, and Instagram Reels all work through a single input field. The app detects the platform and routes through the correct pipeline automatically — the user never thinks about it.
A complete product, not a demo. Preplo isn't a proof of concept — it has authentication, subscriptions, premium features, push notifications, a polished landing page, and a code-generated demo video. We built something we'd actually ship.
Cook mode that works in a real kitchen. We tested this while actually cooking. Screen stays on, text is huge, swipe between steps, tap to hear instructions read aloud. It's designed for flour-covered hands, not a demo booth.
The economics work at scale. Our caching architecture means every video that gets extracted once is free forever. The more users join, the faster and cheaper Preplo gets — a rare case where growth improves margins instead of hurting them.
We went from idea to App Store-ready in a hackathon. The scope of this project — backend, AI pipeline, mobile app, engagement system, landing page, demo video — is something that usually takes months. We're proud we pulled it together and made every piece feel finished.

What we learned

Start with the problem, not the technology. We could have built a generic "AI wrapper" app, but we kept coming back to one question: would we actually use this tonight? That filter killed half our feature ideas and made the surviving ones 10x better.
AI is only as good as the system around it. The model itself is powerful, but without validation, caching, and error handling around it, the output isn't reliable enough to cook from. Making AI trustworthy is harder than making it work.
Design for the kitchen, not the screen. Cook mode went through three redesigns once we started testing while actually cooking. Font sizes that looked fine at a desk were unreadable at arm's length next to a stove. Real-world context changed everything.
Gamification needs empathy. Our first streak system was too strict and felt punishing. Once we added freezes and switched to weekly cadence, it stopped feeling like a chore and started feeling like encouragement. The mechanic is the same — the emotion is completely different.
Scope is the hardest design decision. We said no to meal planning, social features, and community feeds — not because they're bad ideas, but because shipping a polished core experience matters more than shipping everything half-done.

What's next for Preplo

Meal Planning — Select recipes for the week, auto-generate a merged shopping list, and get daily cooking reminders with smart scheduling.
Social Cooking — Share recipes with friends, see each other's streaks, and build cooking accountability together.
Voice Control in Cook Mode — "Next step," "repeat that," "set a timer for 5 minutes" — fully hands-free cooking powered by voice.
Grocery Integration & Ordering — Real-time pricing from nearby stores, and the ability to order ingredients directly from within the app. Go from video to doorstep without leaving Preplo.
Creator Dashboard — A dedicated space for content creators to see how their videos are performing on Preplo — which recipes are being extracted, cooked, and saved the most, audience demographics, and engagement trends. Turning creators from content sources into partners.
Android Launch — Our cross-platform architecture means Android is already in progress.
Community & Collections — Curated recipe collections and a community feed of the best extracted recipes from across the platform.