LeftoverLab

A privacy-first iOS kitchen assistant that turns what you already have into meals you can actually make.

LeftoverLab answers one everyday question — "I have food at home, but I still don't know what to eat." You add your ingredients, leftovers, and ready-to-cook items, and the app ranks the recipes you can make right now, what you're one ingredient short of, and how to use up food before it expires. Everything runs on-device: no account, no backend, and no data leaves your phone. (The only outbound links are optional taps to a recipe creator's YouTube video.)

---

Why it exists

The real problem isn't a lack of food — it's decision fatigue plus poor visibility into what's already in the kitchen. People own enough to make a meal but default to ordering out, then let food expire. LeftoverLab turns existing inventory into a short list of concrete, makeable meals in seconds, and nudges you to use leftovers and soon-to-expire items first.

---

Screenshots

Use Your Leftovers	Ranked Suggestions	Kitchen Inventory
Add Items	Preferences	Cuisine & Creators

---

Features

• Smart inventory — track ingredients, leftovers, and ready-to-cook items, grouped and collapsible, with optional expiration dates.
• Ranked suggestions — a deterministic matcher sorts recipes into Quick, Can Make Now, Uses Expiring Soon, and Missing One Ingredient, and tells you why each was suggested.
• Leftover intelligence — cooked leftovers surface in their own section with dish-aware pairing (e.g. dal with rice, sabzi with rice or roti) and a "use the base you already have, otherwise cook a little rice" nudge — only at lunch and dinner.
• Smart ingredient chips — surfaces the ingredients that would unlock the most new recipes; tap to add and re-rank instantly.
• Mealtime + cuisine + diet filters — auto-selects the current mealtime, with cuisine, diet (vegetarian / egg / chicken / other meat), and nutrition filters.
• Ingredient normalization — matches plurals and English/Hindi synonyms, so "curd" finds yogurt recipes and "roti" matches tortilla.
• Recipe detail + hands-free Cooking Mode — one large step at a time with spoken narration for cooking without touching the screen.
• Receipt scanning — add multiple items at once from a photographed receipt using on-device Vision OCR.
• Accessibility first — VoiceOver labels, Dynamic Type, and large tap targets throughout.

---

Architecture

LeftoverLab uses SwiftUI + SwiftData with a small, pure services layer, deliberately separating concerns so the core logic is testable in isolation.

        Views (SwiftUI)  ──▶  Services
              ▲                  │
              │                  ├── RecipeMatcher        (pure scoring + classification)
   SwiftData @Model:             ├── RecipeCatalog        (loads bundled recipes.json)
   InventoryItem                 ├── IngredientNormalizer (name matching / synonyms)
              ▲                  └── LeftoverSuggester    (dish-aware leftover pairing)
              │
   recipes.json (bundled, read-only)

Key design decisions:

• Two data lifetimes. Mutable user data (inventory) lives in SwiftData; the recipe catalog is immutable content loaded from a bundled JSON. They're never conflated.
• Classification is decoupled from scoring. Which bucket a recipe lands in is driven by how many required ingredients are missing — not by its score — so a "Can Make Now" label never lies.
• Coverage as a ratio. Scoring normalizes by recipe size (ownedRequired / totalRequired), so a fully-makeable small recipe beats a half-stocked large one.
• The matcher is a pure function. (inventory, recipes) -> [MealSuggestion] with no UI or persistence inside it, which makes it fast, explainable, and unit-testable.

---

Testing

The matching engine is covered by a Swift Testing suite (MatcherTests.swift) spanning bucket classification, pantry-staple exclusion, plural/synonym normalization, the rice-or-roti base rule, score ordering, expiring-item logic, and the smart-ingredient suggestions. Because the matcher is decoupled from UI and SwiftData, the suite runs in milliseconds and guards against regressions when the ranking logic changes.

Run with ⌘U in Xcode.

---

Tech stack

Swift · SwiftUI · SwiftData · Swift Testing · Vision (OCR) · AVFoundation (speech)

Requirements

• Xcode 26+
• iOS 26+
• iPhone (tested on the iPhone 17 Pro simulator)

Getting started

1. Clone the repo and open LeftoverLab.xcodeproj in Xcode.
2. Make sure recipes.json is included in the app target's Copy Bundle Resources.
3. Select an iPhone simulator and run (⌘R).

---

Roadmap

• On-device AI recipe generation (Apple Foundation Models) for ingredients the catalog doesn't cover
• App Intents / Siri ("What can I cook right now?")
• Expiration notifications and a waste/savings summary

---

Note on the recipe catalog

The bundled catalog is a curated set of original recipes, with a focus on Indian home cooking plus some Western dishes. It's intentionally finite and ships with the app so everything works offline.