Introduction

Hi, I'm Hazuki — an indie developer based in Japan.

I built and shipped Smart Inventory, a family grocery management app, using Flutter, Firebase, and the Gemini AI API — entirely from a Windows machine, without ever touching a Mac.

In this post, I'll walk through the challenges I ran into and how I solved them. If you're in a similar situation, I hope this helps.

What I Built

Smart Inventory — Family Grocery & Home Tracker

A real-time inventory and shopping list sharing app designed for busy, dual-income families.

Key Features

• Inventory Management: Organize food and household items by category (fridge, freezer, pantry, spices, daily essentials)
• Expiry Alerts: Color-coded warnings for items expiring soon (🔴 today / 🟠 within 3 days)
• Shared Shopping List: Real-time sync across the family — shows who added what
• AI Recipe Suggestions: Gemini AI suggests personalized recipes based on what's currently in stock
• Household Profile: Recipes are personalized based on allergies, cooking time, skill level, and cuisine preferences
• Premium Subscription: Applied at the household level — one purchase covers the whole family ### Tech Stack

Why I Built It

It started with small but persistent frustrations of dual-income household life:

• The "bought it again" problem: Three bottles of soy sauce. Two packs of eggs.
• The "what's in the fridge?" problem: Having to text your partner before every grocery run
• The "what's for dinner?" problem: Coming home exhausted and having no idea what to cook I tried existing inventory apps, but none combined real-time family sharing with AI-powered recipe suggestions in one place. So I decided to build it myself.

Technical Challenges & Solutions

1. Building an iOS App Without a Mac

Buying a Mac just for indie development is a big investment. After some research, I found Codemagic — a cloud CI/CD service that builds on real Mac hardware in the cloud.

# codemagic.yaml overview
workflows:
  ios-release:
    name: iOS Release
    instance_type: mac_mini_m2
    environment:
      ios_signing:
        distribution_type: app_store
        bundle_identifier: com.hazuki-tech.smartInventory

Codemagic handles the build on a cloud Mac, so everything stays on Windows. Certificate management is handled automatically using an App Store Connect API key — no Mac required at any step.

2. Provisioning Profile Missing In-App Purchase Entitlement

After adding IAP support, my builds started failing with this error:

Error: Provisioning profile doesn't include 
the com.apple.developer.in-app-payments entitlement.

Root cause: The In-App Purchase capability wasn't enabled on the App ID in Apple Developer.

Fix:

1. Apple Developer → Identifiers → Select your App ID
2. Enable In-App Purchase under Capabilities
3. Regenerate the provisioning profile
4. Set Codemagic Code Signing to "Automatic" ### 3. RevenueCat Webhook Bug: CANCELLATION vs EXPIRATION

I had a critical bug in my subscription webhook logic.

Problem: I was treating CANCELLATION and EXPIRATION the same way, which caused users to lose Premium access immediately after cancelling — even though they'd paid for the rest of the billing period.

Correct behavior:

• CANCELLATION = User cancelled, but they're still in the paid period → keep Premium
• EXPIRATION = Billing period actually ended → downgrade to Free

case 'CANCELLATION':
  // Still within paid period — keep Premium
  await updateSubscriptionStatus(householdId, {
    plan_type: 'premium',
    premium_expiry: expirationAtMs,
  });
  break;

case 'EXPIRATION':
  // Billing period ended — downgrade to Free
  await updateSubscriptionStatus(householdId, {
    plan_type: 'free',
    premium_expiry: null,
  });
  break;

4. Gemini API Ignoring Allergy Constraints

Even when I included allergy restrictions in the prompt, Gemini occasionally suggested recipes containing those ingredients.

Root cause: The allergy constraint was buried among other conditions with no special priority signaling.

Fix: Move the constraint to the very top of the prompt as a standalone section with strong emphasis.

⚠️ ABSOLUTE PRIORITY — MUST FOLLOW WITHOUT EXCEPTION:
The following dietary restrictions must NEVER appear in any recipe,
even in small amounts, as hidden ingredients, or in alternative forms.
This rule overrides all other instructions.

[FORBIDDEN INGREDIENTS]
- Eggs and egg-based products (mayonnaise, egg yolk, egg white, etc.)

I'm using gemini-2.5-flash as the primary model with gemini-2.5-pro as a fallback, with a 120-second timeout to handle more complex recipe generation.

5. Riverpod Late Initialization Bug

User preferences from the household profile weren't being reflected in Gemini's recipe suggestions.

Root cause: RecipeRepository was being instantiated before HouseholdProfileProvider had finished loading from Firestore, resulting in profile: null.

Fix: Explicitly await the Firestore data before accessing the repository.

// Before: profile is null when RecipeRepository is created
final repository = ref.read(recipeRepositoryProvider);

// After: wait for Firestore data before accessing the repository
await ref.read(householdProfileProvider.future);
final repository = ref.read(recipeRepositoryProvider);

6. Preventing Duplicate Recipe Suggestions

Gemini was occasionally suggesting the same recipes repeatedly.

Fix: Fetch the 5 most recent recipe titles from Firestore and inject them into the prompt as a strict prohibition list.

const recentRecipes = await getRecentRecipes(householdId, 5);
const prohibitedList = recentRecipes.map(r => `- ${r.title}`).join('\n');

const prompt = `
STRICT RULE — DO NOT SUGGEST:
The following recipes (and anything similar) must not be suggested:
${prohibitedList}
`;

Recipe Usage Limits

Since AI recipe generation calls the Gemini API via Cloud Functions, cost control matters. Usage is tracked server-side in Firestore to prevent client-side manipulation.

The counter is only incremented after a successful API response — failed requests don't count against the limit.

Dev Tools

The Journey to Launch

1. Planning & Design: Requirements, Firestore schema design
2. Core Features: Inventory, shopping list, Firebase integration
3. AI Features: Gemini API proxied through Cloud Functions
4. Payments: RevenueCat integration, Webhook setup
5. CI/CD: Codemagic pipeline, automated builds
6. App Store Prep: Screenshots, descriptions, privacy policy

7. Submission: Review approved → manual release

Lessons Learned

What went well

• No Mac, no problem: Codemagic removed the biggest barrier to iOS development on Windows
• Firebase + Riverpod is a great combo: StreamProvider made real-time sync surprisingly easy

• Claude Code accelerated development significantly: Passing a CLAUDE.md spec file and letting it handle complex implementations saved a huge amount of time
  What was hard

• RevenueCat subscription lifecycle: The CANCELLATION vs EXPIRATION distinction caught me off guard

• Prompt engineering: Getting an LLM to reliably follow strict constraints is harder than it looks

- IAP testing without a Mac: Sandbox testing for in-app purchases is tricky without direct device access

Closing Thoughts

You don't need a Mac to ship an iOS app as an indie developer. With tools like Codemagic, the barrier is much lower than it used to be.

Smart Inventory is live on the App Store — search for "Smart Inventory" or grab it at the link below. Feedback is always welcome!

https://apps.apple.com/us/app/smart-inventory-family-pantry/id6760807269

Feel free to reach out on X: @hazuki_tech_dev

If this was useful, a like or comment goes a long way 🙏