DressApp

Your AI fashion editor: photograph any garment, get weather + calendar-aware outfits, scan EU DPP QR tags, and resell from a polished marketplace — all in your pocket.

Live demos

• 🟢 Production · https://dressapp.co (Hetzner VPS, custom domain)
• 🚀 Contest deployment · https://ai-stylist-api.emergent.host (Emergent native)

---

What it does

DressApp turns a closet of physical clothes into a structured, queryable wardrobe and uses that wardrobe to drive everyday styling decisions.

Core flows

---

Tech stack

Backend — FastAPI (Python 3.11) · Motor async MongoDB driver · Pydantic v2

Frontend — React 19 · React Router · Tailwind · Shadcn/UI · react-i18next (12 locales) · Sonner toasts · Lucide icons

Vision pipeline — environment-aware:

• Hetzner / dev (full stack): rembg (U2-Net) for matting · HuggingFace SegFormer-b2-clothes for clothing parsing · Fashion-CLIP for embeddings — all CPU-local.
• Emergent host (lightweight pod, 250 m CPU / 1 Gi RAM): Gemini Nano Banana for multi-item detection · HuggingFace Inference API for SegFormer · Cleanly skips matting when rembg is absent.
• The same image runs on both — see requirements-ml.txt and the auto-detection in app/config.py.

Voice — Deepgram (STT + TTS, multi-voice)

LLM — Direct GEMINI_API_KEY in production · Emergent universal key in dev. Default stylist model: Gemini Flash 2.x. Image generation: Gemini 2.5 Flash Image (Nano Banana).

External APIs — OpenWeather · PayPal Live · Google OAuth + Google Calendar · HuggingFace Inference API

Hosting — Docker Compose · Caddy 2 (auto Let's Encrypt) · MongoDB Atlas

---

Architecture at a glance

                  ┌──────────── HTTPS ────────────┐
                  │                                │
        ┌─────────▼────────┐              ┌────────▼─────────┐
        │  Caddy (TLS)     │              │ Browser (PWA)    │
        │  /api/* → backend│              └──────────────────┘
        │  /*    → frontend│
        └────┬────────┬────┘
             │        │
   ┌─────────▼┐    ┌──▼─────────────────┐
   │ FastAPI  │    │ Nginx (static SPA) │
   │  :8001   │    │  React build       │
   └────┬─────┘    └────────────────────┘
        │
        ├─ MongoDB Atlas (users, closet, listings, trends, …)
        ├─ Vision pipeline · auto-selected per deploy:
        │    · Hetzner   → local SegFormer + rembg + Fashion-CLIP
        │    · Emergent  → Gemini Nano Banana detector + skip-matting
        ├─ Deepgram (STT/TTS over HTTPS)
        ├─ OpenWeather, PayPal Live, Google OAuth/Calendar
        └─ Direct GEMINI_API_KEY (prod) or Emergent LLM key (dev)
             → text + Nano Banana image generation

A more detailed write-up lives in /app/docs/ARCHITECTURE.md and the database shape in /app/docs/MONGODB_SCHEMA.md.

---

Repository layout

.
├── backend/                # FastAPI service
│   ├── server.py           # App entry, ASGI bindings, CORS
│   ├── app/
│   │   ├── api/v1/         # Versioned routers (closet, listings, stylist, …)
│   │   ├── services/       # Business logic + ML services
│   │   ├── models/         # Pydantic schemas
│   │   ├── db/             # Mongo bootstrap + index creation
│   │   └── core/           # Settings, security, deps
│   ├── scripts/
│   │   └── seed_demo.py    # Idempotent demo-data seeder
│   ├── requirements.txt    # Lightweight deps — installed by both deploys
│   └── requirements-ml.txt # Heavy ML stack (torch, transformers, rembg)
│                           # — only installed by the Hetzner Dockerfile
│
├── frontend/               # React SPA (Create-React-App + craco)
│   ├── src/
│   │   ├── pages/          # Top-level routes
│   │   ├── components/     # Shared UI (DppScanner, AdTicker, …)
│   │   ├── components/ui/  # Shadcn primitives
│   │   ├── lib/            # api client, i18n, helpers
│   │   └── locales/        # 12 translation files
│   └── package.json
│
├── deploy/                 # Production deploy kit
│   ├── docker-compose.yml
│   ├── Dockerfile.backend
│   ├── Dockerfile.frontend
│   ├── Caddyfile
│   ├── nginx-frontend.conf
│   └── DEPLOY.md           # Step-by-step VPS guide
│
└── docs/                   # Architecture + schema docs

---

Getting started — local dev

The whole project ships ready to run inside the Emergent platform. Backend, frontend and MongoDB are already wired up by supervisord.

Backend

cd backend
# Hot-reload is on in supervisor; you only need to restart on dep changes:
sudo supervisorctl restart backend
# Required env (already populated in /app/backend/.env):
#   MONGO_URL, DB_NAME, JWT_SECRET, EMERGENT_LLM_KEY
#   GOOGLE_OAUTH_CLIENT_ID/SECRET, DEEPGRAM_API_KEY, OPENWEATHER_API_KEY,
#   GROQ_API_KEY, HF_TOKEN, PAYPAL_LIVE_CLIENT_ID/SECRET, PAYPAL_ENV

Frontend

cd frontend
yarn install
sudo supervisorctl restart frontend
# REACT_APP_BACKEND_URL is preconfigured in frontend/.env

Seed demo data (optional)

For a fresh database, populate listings, professionals, trend cards and a demo user:

cd backend
python -m scripts.seed_demo   # idempotent — re-running upserts

---

Production deployment

DressApp ships one codebase, two production targets that share the same backend image. The vision pipeline auto-detects which dependencies are present and chooses a code path accordingly — no per-deploy env overrides needed.

Target A — Hetzner VPS (dressapp.co) · full local ML

3-container Docker Compose stack on any 4 GB+ VPS:

• backend — FastAPI + local SegFormer + rembg + Fashion-CLIP (~1.5 GB RAM at idle)
• frontend — Nginx serving the built SPA
• caddy — TLS termination, automatic Let's Encrypt, HTTP→HTTPS redirect

deploy/Dockerfile.backend installs requirements.txt and requirements-ml.txt so torch / transformers / rembg are all present. app/config.py auto-detects them and turns USE_LOCAL_CLOTHING_PARSER and AUTO_MATTE_CROPS to true.

Step-by-step instructions: deploy/DEPLOY.md.

# After cloning on the VPS:
cd deploy
cp .env.example .env   # then edit
docker compose up -d --build

Routine update:

git pull origin main
docker compose up -d --build

Target B — Emergent host (ai-stylist-api.emergent.host) · cloud-only ML

Emergent's auto-deploy pod is sized at 250 m CPU / 1 Gi RAM, which cannot host the local ML stack. The deploy pipeline only installs backend/requirements.txt (no torch / transformers / rembg / cuda-*), so:

• app/config._HAS_TORCH / _HAS_REMBG resolve to False.
• USE_LOCAL_CLOTHING_PARSER and AUTO_MATTE_CROPS default to false.
• The analyse pipeline falls through to the Gemini multi-item detector (already implemented in garment_vision._gemini_detect), and matting cleanly returns None so the original crop is kept.

If Emergent's build cache still ships rembg after requirements.txt removed it, the auto-detect will keep the heavy paths enabled and the pod will start hanging on every analyse call (180 s rembg model download + 2 K image inference exceeds the gateway timeout). To force-disable both heavy paths regardless of installed wheels, set LIGHTWEIGHT_DEPLOY=true in Emergent's env dashboard. This single override pins AUTO_MATTE_CROPS=false and USE_LOCAL_CLOTHING_PARSER=false at boot.

A health probe at GET /api/v1/closet/analyze/version exposes the active mode:

// dressapp.co
{ "torch_installed": true,  "use_local_clothing_parser": true,  "auto_matte_crops": true,  "lightweight_deploy": false, ... }
// ai-stylist-api.emergent.host  (with LIGHTWEIGHT_DEPLOY=true)
{ "torch_installed": false, "use_local_clothing_parser": false, "auto_matte_crops": false, "lightweight_deploy": true,  ... }

The probe is fast by default (skips the live rembg matting cycle so it never times out behind a 60 s gateway). Pass ?probe=1 when you want the heavier health check.

Both targets serve identical user-facing functionality — Add Item, Reanalyse, Clean Background, Stylist, Marketplace — the only difference is where the ML runs.

---

Contributing

1. Fork → branch off main
2. Run lint before pushing:
   • Python — ruff check backend/
   • JS/TS — cd frontend && yarn lint
3. Open a PR. CI runs the full test suite (FastAPI tests + Playwright smokes).

---

License

© 2025–2026 DressApp. All rights reserved. The trademarks, product designs, and content displayed in the app are the property of their respective owners.