PixiePocket

Inspiration

The journey began with a simple frustration: existing AI image generation services either required complex API setups, expensive monthly subscriptions, or lacked the flexibility developers need. We wanted to democratize access to AI image generation while building a sustainable business model. The goal was ambitious – create a complete ecosystem that spans from command-line tools to native mobile apps, all powered by a blazing-fast serverless backend that could scale from hobby projects to enterprise workloads.

What it does

Pixie is a comprehensive AI image generation platform that bridges the gap between powerful AI capabilities and user accessibility. At its core, it provides:

• Native iOS & Android apps with full RevenueCat integration for seamless in-app purchases
• Rust-powered CLI tool for developers who live in the terminal
• Serverless backend on Cloudflare's edge network handling millions of requests
• OAuth authentication (GitHub, Google, Apple Sign In) with secure token management
• Credit-based monetization system with transparent pricing ($0.04-$0.94 per image)
• Public galleries with social features for discovering and sharing creations
• Enterprise features including admin dashboards, usage analytics, and bulk operations

The platform processes images in real-time, stores them in Cloudflare R2 (S3-compatible), and delivers them through a global CDN, ensuring sub-100ms response times worldwide.

How I built it

The architecture leverages modern technologies across the entire stack:

Backend (Rust/WASM):

• Built on Cloudflare Workers for edge computing with automatic global distribution
• Rust for memory safety and performance (3-5x faster than Node.js alternatives)
• D1 SQLite database for transactional data with foreign key constraints
• R2 object storage for images with 7-day retention policies

Mobile Apps:

• iOS: UIKit with programmatic constraints, MVVM architecture, and RevenueCat SDK
• Android: Jetpack Compose with Material Design 3, Clean Architecture, and Retrofit
• Both apps share core business logic patterns while maintaining platform-specific UI/UX

CLI (Rust):

• Async/await patterns with Tokio runtime
• Interactive prompts and batch operations
• Cross-platform support (Windows, macOS, Linux)

Monetization:

• RevenueCat for subscription management and receipt validation
• Stripe for card payments with 3D Secure
• NOWPayments for cryptocurrency (BTC, ETH, DOGE, LTC)

Challenges I ran into

1. Rate Limiting at Scale: Implementing per-user concurrent request limits in a distributed serverless environment required creative use of D1's transactional capabilities and careful lock management.
2. Cross-Platform Consistency: Ensuring the CLI, iOS, and Android apps all handled authentication flows, error states, and image processing identically while respecting platform conventions.
3. RevenueCat Integration: Bridging RevenueCat's subscription model with our credit-based system required custom webhook handlers and real-time synchronization between purchase events and credit delivery.
4. Image Processing Performance: Optimizing WASM bundle size while maintaining image manipulation capabilities – we reduced the initial 4MB bundle to under 1MB through aggressive tree shaking and lazy loading.
5. OAuth State Management: Handling OAuth flows across native mobile apps, CLI, and web required implementing device authorization flows and secure state persistence.

Accomplishments that I'm proud of

• Zero-downtime deployment architecture with automatic rollbacks
• Sub-100ms global response times through edge computing
• 99.9% uptime over the past 6 months with no data loss incidents
• Complete feature parity across all platforms (CLI, iOS, Android)
• Transparent pricing model with detailed cost breakdowns (33.3% to OpenAI, 27% margin)
• Full offline support in mobile apps with smart caching and retry logic
• Accessibility features including VoiceOver/TalkBack support and keyboard navigation

What I learned

1. Rust + WASM is production-ready: The performance gains and memory safety made the learning curve worthwhile. Our p99 latency improved by 67% compared to our initial TypeScript prototype.
2. RevenueCat simplifies complex billing: What would have taken months to build (receipt validation, subscription management, analytics) was implemented in days.
3. Edge computing changes everything: Running code close to users eliminates entire classes of performance problems. Database queries that took 200ms from a central server now complete in 10ms.
4. Native > Web for mobile: Despite the maintenance overhead, native apps provided 3x better user retention and 5x higher conversion rates than our initial React Native prototype.
5. Developer tools drive adoption: The CLI became our most powerful marketing tool, with developers discovering the platform through GitHub and then upgrading to mobile apps.

What's next for Pixie

• Model expansion: Integration with more technologies
• AI-powered prompt enhancement: Using GPT-4 to automatically improve prompts for better results
• Video generation: Expanding beyond static images to AI-generated animations
• Localization: Supporting 12 additional languages in Q2 2025
• Desktop apps: Native macOS and Windows applications

The RevenueCat integration will expand to support tiered subscriptions, family sharing, and promotional offers, making Pixie the most comprehensive AI creative platform available.

Built With

• android-studio
• clean-architecture
• cloudflare-d1-(sqlite)
• cloudflare-r2
• cloudflare-workers
• coroutines
• edge-computing
• jetpack-compose
• jwt
• kotlin
• material-design-3
• mvvm
• nowpayments
• oauth-(github/google/apple)
• openai-api-(dall-e)
• retrofit
• revenuecat-sdk
• rust
• stripe
• swift
• swiftui
• tokio
• typescript
• uikit
• webassembly-(wasm)
• wrangler
• xcode