CLO: from zero to 4,000+ users in 5 months.
- Product
- Fashion social network with an AI stylist
- Platforms
- App Store and Google Play
- My role
- Co-founder and sole developer
- Period
- Jan 2026 – present
The challenge
CLO needed an app where each person builds a virtual closet from their own clothes and gets outfit suggestions from AI. The first version, a WebView, stuttered in the feed and took about 3 seconds to load. The AI had to analyze thousands of photos without failing when the provider rate-limited requests, and without blowing up the cost per subscriber.
What was built
- Clothing analysis by AI that doesn’t fall overEvery photo is processed in the background, with automatic retries when the AI provider limits volume and without ever processing the same item twice.
- An assistant that proposes before it actsThe assistant suggests the action and only runs it after the user confirms, with usage limits per subscription plan.
- A faster, cheaper backendServer and database changes took 1.2 s off every request and cut database round-trips from 5 to 3.
- A smooth native app and paymentsThe feed was rewritten in Flutter to run at 60 frames per second, with subscriptions integrated with the Apple and Google stores.
Technical details
See the engineering pageMultimodal pipeline
Gemini / Vertex AI via Hangfire; per-item embeddings and Lab palettes; idempotent jobs; backoff on 429/503.
Parallel function calling
Propose → confirm flow, double-execution protection and token quotas per tier.
.NET 8 + EF Core
−1.2 s per request on connection startup; BLOBs out of hot queries (5 → 3 round-trips).
Infrastructure and payments
Cloud Run + Cloud SQL over Direct VPC; idempotent RevenueCat webhooks; Flutter replacing the Ionic WebView.