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 demo

• 🟢 Production · https://dressapp.co (Hetzner VPS, custom domain)

---

What it does

DressApp turns a closet of physical clothes into a structured, quarriable 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 · useSyncExternalStore custom stores · React Router · Tailwind · Shadcn/UI · react-i18next (13 locales) · Sonner toasts · Lucide icons

Vision pipeline — full stack:

• Hetzner / dev: rembg (U2-Net) for matting · HuggingFace SegFormer-b2-clothes for clothing parsing · Fashion-CLIP for embeddings — all CPU-local. See requirements-ml.txt and the auto-detection in app/config.py.
• Inference Server: Optional self-hosted GPU endpoint (Modal/RunPod) wrapping SegFormer-b3 and BiRefNet for fast background matting and parsing.

Voice / Audio — Deepgram (STT/TTS fallback), Gemini 2.5 Flash native modulations (prebuilt voice configs puck/aoede/charon), Web Speech API browser integration, and local Piper ONNX support.

LLM / Inference Engine — Multi-tier production routing:

• Primary Production Engine (Google Gemini 3.5 Flash-Lite): Powers all 6 interactive and core pipelines (Stylist Brain, Wardrobe Migration, Suitcase Planner, Trend Scout localization, Session Titles, and Closet Ingestion) via backend/app/services/llm_gateway.py with native google-genai SDK and sub-350ms TTFT.
• On-Premises VPS Eyes (gemma-4-E4B) — Free Tier & Quota Safety Net: Self-hosted dressapp-eyes container running fine-tuned Gemma-4-E4B (gemma-4-E4B-it-Q3_K_M.gguf via llama-server on port 7860). Serves as the zero-cost baseline for offline/free usage and transparently intercepts quota exhaustion (429 / RESOURCE_EXHAUSTED) with zero downtime.
• High-Cost Generative Features: Trend Scout and Nano Banana photorealistic inpainting strictly require validated user keys.

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

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

Extension — Manifest V3 Shopping Assistant (React popup + content scripts)

---

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: local SegFormer + rembg + Fashion-CLIP
        ├─ Primary Engine: Google Gemini 3.5 Flash-Lite (native google-genai)
        ├─ On-prem Eyes (:7860): llama-server + Gemma-4-E4B (Free Tier & Quota Safety Net)
        ├─ Deepgram & Gemini Audio (STT/TTS over HTTPS)
        ├─ OpenWeather, PayPal Live, Google OAuth/Calendar
        └─ High-cost generative cloud endpoints (BYOK keys for Nano Banana & Trend Scout)

In-depth documentation

• Technical Architecture: docs/ARCHITECTURE.md — Monorepo structure, FARM runtime, ML vision stack, and Hetzner VPS topology.
• Database Schema: docs/MONGODB_SCHEMA.md — Complete 15-collection MongoDB Atlas schema, indexes, and credit bucket definitions.
• Domain Context & ADRs: CONTEXT.md and docs/adr/ — Bounded contexts, architectural decision records, and deep module boundaries.
• Modular Wiki: wiki/en/overview.md — 40 detailed feature guides localized into 13 languages.
• Deployment Facts: CONCRETE_FACTS.md — Authoritative hardware, container layouts, and operational reference commands.

---

Repository layout

.
├── apps/
│   ├── web/                     # React 19 Web SPA (CRA + Craco, Tailwind, i18next)
│   ├── mobile/                  # Expo 53 / React Native 0.79 Mobile App (iOS / Android)
│   └── android-twa/             # Android Trusted Web Activity packaging wrapper
│
├── packages/
│   ├── api-client/              # Shared isomorphic REST & streaming NDJSON client
│   ├── i18n/                    # Canonical translation catalogs across 13 locales
│   └── types/                   # Shared TypeScript definitions for domain models
│
├── backend/                     # FastAPI Python 3.11 Backend Service
│   ├── server.py                # ASGI application root & CORS configuration
│   ├── app/
│   │   ├── api/v1/              # Versioned API routes (auth, closet, stylist, marketplace, etc.)
│   │   ├── core/                # Configuration, JWT security, and dependency injection
│   │   ├── db/                  # Motor MongoDB client and index initialization
│   │   ├── models/              # Pydantic v2 schemas (schemas.py, credit.py)
│   │   └── services/            # Business logic, vision matting, stylist agent, and billing
│   ├── scripts/                 # Seeding and maintenance utilities
│   ├── requirements.txt         # Production backend dependencies
│   └── requirements-ml.txt     # Local ML stack (torch, transformers, rembg)
│
├── chrome-extension/            # Manifest V3 Shopping Assistant Extension
├── inference-server/            # Optional standalone GPU inference server (BiRefNet / Gemma-4)
├── deploy/                      # Production Docker Compose, Caddyfile, and Dockerfiles
│   ├── docker-compose.yml
│   ├── Dockerfile.backend
│   ├── Dockerfile.frontend
│   ├── Caddyfile
│   ├── nginx-frontend.conf
│   └── DEPLOY.md                # Step-by-step VPS deployment guide
├── docs/                        # Technical documentation, ADRs, and schema references
│   ├── ARCHITECTURE.md          # Full system architecture
│   ├── MONGODB_SCHEMA.md        # MongoDB Atlas collection schemas
│   ├── adr/                     # Architectural Decision Records (0001 - 0004)
│   └── agents/                  # AI agent guidelines and domain documentation
├── .agents/                     # AI agent skills & workflows
│   └── workFlows/               # Core workflows (deploy, labrrerian, translator, user-manual)
├── quarantine/                  # Archived historical session notes and debug logs
└── wiki/                        # Localized two-layer user documentation (13 languages)

---

Getting started — local dev

Monorepo Setup

# Install workspace dependencies from the repo root:
yarn install

Backend

cd backend

# Create and activate a Python 3.11 virtual environment:
python -m venv .venv
# On Linux/macOS:
source .venv/bin/activate
# On Windows PowerShell:
.\.venv\Scripts\Activate.ps1

# Install dependencies:
pip install -r requirements.txt

# Run the FastAPI development server:
python server.py   # Or: uvicorn server:app --reload --port 8001

Required environment variables in backend/.env: MONGO_URL, DB_NAME, JWT_SECRET, GEMINI_API_KEY, GOOGLE_OAUTH_CLIENT_ID, GOOGLE_OAUTH_CLIENT_SECRET, DEEPGRAM_API_KEY, OPENWEATHER_API_KEY, PAYPAL_LIVE_CLIENT_ID, PAYPAL_LIVE_CLIENT_SECRET, PAYPAL_ENV.

Web Frontend

# Start the React 19 web application:
yarn --cwd apps/web start
# Access at http://localhost:3000 (REACT_APP_BACKEND_URL defaults to http://localhost:8001)

Mobile App (Expo)

# Start the Expo development server:
yarn --cwd apps/mobile start

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

4-container Docker Compose stack on Hetzner Cloud CPX32 VPS (4 AMD vCPUs, 8 GB RAM, running on dressapp.co):

• backend — FastAPI + local SegFormer + rembg + Fashion-CLIP (~1.5 GB RAM at idle)
• eyes — Self-hosted inference server (llama-server + FastAPI proxy) running fine-tuned gemma-4-E4B-it-Q3_K_M.gguf on port 7860 (~2.85 GB RAM)
• frontend — Nginx serving the built React 19 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

---

Contributing

1. Fork → branch off main
2. Run lint before pushing:
   • Python — ruff check backend/
   • TypeScript / Web — yarn --cwd apps/web lint
   • Mobile — yarn --cwd apps/mobile lint
3. Open a PR. CI runs the full test suite.

---

License

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