Sous - AI Recipe Keeper

Turn any recipe - from cookbooks to food videos - into a smart grocery list, instantly. Stop saving recipes. Start cooking them.

Comment

Inspiration

Eitan Bernath said it best in his brief: "Getting from a recipe you want to make to actually cooking it in your kitchen is a real challenge."

That hit home hard. Most of us live in two parallel universes when it comes to food:

  • A fantasy universe - overflowing with saved Instagram Reels, bookmarked YouTube videos, dog-eared cookbook pages, and screenshots of recipes we swear we'll make someday.
  • A reality universe - staring at the fridge on a Wednesday night, ordering takeout again.

The gap between those two worlds isn't a lack of inspiration. There's more food content being created today than at any point in human history. The gap is friction.

Friction of extracting a recipe from a 12-minute video. Friction of cross-referencing three cookbooks against what's actually in your pantry. Friction of building a grocery list from scratch every single time. Friction that quietly convinces you it's just easier not to cook.

Sous was built to eliminate that friction entirely to be the AI-powered prep work that happens before you even step into the kitchen, so that when you do, you're ready.

What it does

Sous is an AI recipe keeper and grocery planning app that bridges the gap between discovering a recipe and actually cooking it. It works in three powerful ways:

🎥 Recipe Extraction from Anywhere

Paste any URL - a YouTube cooking video, an Instagram Reel, a TikTok, a Pinterest pin, a Facebook post, or any food blog and Sous uses AI to extract the full recipe, ingredient list, and steps automatically. No more scrubbing through a 15-minute video to find out how much butter you need. In under 60 seconds, you have a clean, structured recipe saved to your collection.

📚 Cookbook Library via Photo

Photograph the cover and index pages of any cookbook you own. Sous uses GPT-5 Vision to read the index, identify the book, and build a fully searchable recipe map of your entire physical library - no manual entry, no typing, no copyright issues. Your cookbook shelf goes from a beautiful decoration to a dynamic, query-able recipe engine.

🛒 Smart, Pantry-Aware Grocery Lists

Select any recipe and Sous automatically generates a grocery list for it - deduplicated and pantry-aware. It knows you don't need to buy olive oil if it's already in your pantry. It combines "½ cup flour" from three different recipes into a single line item. It strips out everything you already own and gives you only what you actually need to buy. Meal planning goes from a Sunday chore to a 30-second task.

🔑 Key Features

  • 11 ways to save a recipe — social platforms, camera, AI, text, and blog links
  • URL → Recipe in under 60 seconds via AI extraction
  • Cookbook indexing from photos of your physical books
  • Pantry tracker to log what you already have at home
  • Smart grocery list with auto-deduplication and pantry awareness
  • Live monetization via RevenueCat with Free and Pro tiers

How we built it

Sous is built in Flutter for a native, cross-platform experience on iOS (and Android), with a backend powered entirely by Firebase.

Tech Stack

Mobile Framework - Flutter

Web & Video Scraping - Apify

AI Recipe Extraction & Intelligence - ChatGPT (GPT-5.2)

Backend Functions - Firebase Functions

Storage - Firebase Storage

Monetization - RevenueCat

The 11 Ways to Save a Recipe

One of Sous's most deliberate design decisions was making recipe capture meet users wherever they are. The "Save a Recipe" screen supports:

From Socials - direct import from Instagram, Pinterest, TikTok, Facebook, YouTube, or any Website URL. Apify handles the heavy lifting of scraping video transcripts, captions, and recipe page content, feeding clean raw text into GPT-5.2 for structured extraction.

From Your Camera & Photos - users can import a screenshot or saved photo from their Photo Library, take a live Photo of a Dish, or most powerfully photograph the cover page and index pages of any physical cookbook. GPT-5.2 reads the index, builds a searchable recipe map of that book, and adds it to the user's personal cookbook library no manual entry required.

From AI & Text - users can describe what they want in plain English ("Ask AI for a recipe"), paste raw recipe text directly, or paste a link from any of the top food blogs.

The AI Pipeline

Every recipe import regardless of source flows through the same Firebase Functions pipeline:

$$ \text{Input (URL / Image / Text)} \xrightarrow{\text{Apify / Vision}} \text{Raw Content} \xrightarrow{\text{GPT-5.2}} \text{Structured Recipe JSON} \xrightarrow{\text{Firebase}} \text{Ingredients + Steps + Grocery List + Pantry Match} $$

GPT-5.2 doesn't just extract ingredients, it simultaneously:

  • Parses recipe steps in clean, numbered format
  • Generates a grocery list from the ingredients
  • Cross-references the pantry to flag what the user already owns

Cookbook Indexing via Photo

The cookbook feature is particularly innovative. A user photographs their cookbook's cover and index pages. Firebase Functions passes these images to GPT-5.2 Vision, which:

  1. Identifies the book by its cover
  2. Reads the index to extract recipe names and page numbers
  3. Builds a searchable recipe index stored in Firebase Storage

The user's entire cookbook collection becomes a queryable database without a single line of manual data entry, and without storing any copyrighted recipe content.

Monetization Architecture (RevenueCat)

Feature Free Pro ($2.99/Week)
Social & URL imports 3 Unlimited
Cookbook photo indexing 1 book Unlimited
Grocery list generation ✓ ✓
Pantry tracker 3 ✓
Ask AI for a recipe 3 ✓

Challenges we ran into

🕷️ Scraping at Scale Across Every Platform

Getting reliable recipe data from Instagram, TikTok, YouTube, Pinterest, Facebook, and arbitrary websites is not a single problem - it's eleven different problems. Each platform has its own structure, rate limits, and anti-scraping behavior. Our solution was a multi-scraper architecture built on Apify: if the primary scraper for a source fails or times out, Sous automatically falls back to a secondary scraper without the user ever seeing an error. The result is a system that is resilient by design, not by luck.

⚡ Speed - From Paste to Recipe in Under 60 Seconds

A recipe import tool is only useful if it's fast. Nobody wants to wait 3 minutes for a grocery list. We benchmarked every scraper and every available GPT model across hundreds of test imports, ruthlessly cutting the slowest options at each layer. The result: the full pipeline URL in, structured recipe with ingredients, steps, and grocery list out completes in under 60 seconds. For a workflow that previously took users 10–15 minutes of manual work, that feels instant.

📸 Cookbook Indexing from Imperfect Photos

Asking users to photograph their cookbook index pages introduces an unpredictable variable: photo quality. Blurry shots, bad lighting, pages not fully flat, indexes that span 4–6 pages all of these can degrade GPT-5.2 Vision's ability to read the content accurately. Rather than building fragile image pre-processing to compensate, we made the right product decision: detect low-confidence extractions and prompt the user to re-upload with guidance. The cookbook index can be as long as it needs to be multi-page indexes are handled across sequential uploads, building one unified searchable library progressively.

📖 The Copyright Tightrope

Eitan flagged this himself in his brief, and he was right to. Our approach is deliberate: Sous never stores copyrighted recipe content. The cookbook photo feature extracts only the index, recipe titles and page numbers, not the actual recipes. It turns a user's physical cookbook shelf into a searchable catalog that points back to the book, not a repository that replaces it.

🚀 Shipping a Polished, App Store-Approved App in 4 Weeks

Eleven import methods. AI extraction. Cookbook indexing. Pantry tracking. Smart grocery lists. RevenueCat monetization. Firebase backend. In four weeks. Every day required hard prioritization calls, cutting nice-to-haves and doubling down on the core flows that deliver immediate, undeniable value.

Accomplishments that we're proud of

  • 🚀 Live on the App Store - a real, downloadable, monetized app shipped within the 4-week hackathon window
  • 🔀 11 ways to save a recipe - the most comprehensive recipe capture surface we've seen in any app: 6 social platforms, photo library, live camera, AI generation, text paste, and blog link import
  • ⚡ Sub-60 second full pipeline - from any URL or photo input to a complete, structured recipe with ingredients, steps, and grocery list
  • 🛡️ Resilient multi-scraper architecture - automatic fallback between scrapers means near-zero failed imports for users, regardless of platform
  • 📚 Cookbook indexing via photo - a genuinely novel feature that turns any physical cookbook into a searchable digital library using nothing but the camera on your phone
  • 💳 Live RevenueCat monetization - real subscriptions activated from day one of App Store launch

What we learned

Building Sous taught us that the best apps don't add features, they remove steps.

Every decision came back to one question: does this get the user closer to actually cooking, or does it add friction? That lens forced us to cut aggressively and polish relentlessly.

We learned that resilience is a feature. Users never see our fallback scraper logic fire, they just see a recipe appear. That invisibility is the whole point. The best infrastructure is the kind nobody knows exists.

We learned that AI is only as good as the experience around it. GPT-5.2 is powerful, but what makes Sous feel magical is the journey from paste to structured recipe, the speed, the clean reveal, the grocery list that already knows what's in your pantry. The model is the engine. The experience is the car.

And we learned that shipping beats perfecting. The app that is live on the App Store today with real users, real imports, and real subscriptions coming in is worth infinitely more than the perfect app that never shipped.

What's next for Sous - AI Recipe Keeper

Sous launched as an MVP, but Eitan's vision is much bigger and so is ours.

  • Android launch - expand beyond iOS to reach Eitan's full audience across both platforms
  • Sharing & collaboration - let users share recipe collections and grocery lists with a partner or family members
  • Batch meal planning - select multiple recipes at once and Sous generates a single consolidated, pantry-aware grocery list for the entire week in one tap
  • Meal planning calendar - drag recipes onto a weekly calendar and auto-generate a consolidated, pantry-aware shopping list for the whole week

🌟 Long Term Vision

$$ \text{Recipe Discovery} + \text{Pantry Intelligence} + \text{Grocery Automation} = \text{The end of "I don't know what to cook tonight"} $$

The dream is simple: anyone should be able to watch Eitan's latest video on a Sunday morning and be eating that dish by Sunday evening. No friction. No forgotten ingredients. No abandoned bookmarks.

Sous is the bridge that makes that possible and we're just getting started.

Built for Eitan Bernath's brief at RevenueCat Shipyard 2026.
Sous - your AI sous chef, always ready for prep.

Built With

Share this project:

Updates

Submission history