Independent product · 2022 →
MiModa
An app that collects, organises and retrieves outfits and garments in a digital wardrobe — deriving both the outfit reading and its individual pieces from a single photo.
- Client
- Independent product
- Type
- iOS & Android app
- Period
- August 2022 – present
- Role
- Mobile developer · owner of the entire technical build
- Stack
- React Native · Expo · TypeScript · Redux Toolkit · Realm · Firebase · OpenAI
Context
MiModa lets people collect, organise and retrieve outfits and garments in a digital wardrobe. The product started as Sara Capodiferro's idea; she also designed its UI and UX. I turned it into an iOS and Android app, taking on the entire technical build, the third-party integrations, store publishing and ongoing maintenance.
The app saves photos, classifies outfits and garments with categories and tags, applies filters and suggests looks tied to season and weather. There is no proprietary backend: application state and data live on the client, persisted through Realm, while external services cover image analysis, remote config, weather, telemetry and in-app purchases.
The technical challenge
The core challenge was deriving, from a single photo of a person, both an overall reading of the outfit and the individual garments composing it. A single generative response did not give enough control over placement, classification and result quality.
So I built a pipeline combining image processing, object detection and multimodal analysis. The photo is normalised and optionally background-removed; the system locates the subject and candidate regions, then uses OpenAI and Eden AI to classify outfit and garments against taxonomies defined by the app. Results are validated against the detected regions: items that cannot be tied to trustworthy coordinates are discarded, while valid ones are cropped and turned into structured wardrobe entities.
Making it work on a real phone
The solution was refined through manual trials on varied photos and comparisons across providers, models and prompts. To survive real mobile conditions, I modelled the processing as persistent stages, with controlled concurrency, bounded retries and resumption after suspension or network failure.
Using the AI services requires explicit, versioned consent; without it, a manual path remains available throughout.
Outcome
MiModa is published and maintained on the Apple App Store and Google Play. With no structured marketing launch, it passed 10,000 organic downloads on Google Play; on the Italian App Store it holds a 5.0/5 rating across 8 ratings. Public figures verified in September 2026.
Concept, UI and UX: Sara Capodiferro. Official visuals from mimoda.app and the app's public store pages.