Switch It Up

Switch It Up is a fashion virtual-wardrobe and stylist marketplace prototype.

The app idea:

• users sign in and set size, height, waist, shoe size, and fit preferences
• users upload photos of clothes into a virtual wardrobe
• the app creates a virtual body/lookalike styling canvas
• a production version can use a guided 360 body video to improve avatar sizing
• normal users can ask stylists to remix existing clothes or suggest replacements
• stylists can build credibility through posts, likes, comments, followers, and helped clients
• malls/shops can list clothes, and stylists can send wishlist outfits to users
• competitions let multiple stylists submit outfits for a paid prize

MVP Features

• Production-style responsive frontend shell
• Apple-inspired glass app layout with sidebar, setup progress, virtual styling canvas, and compact controls
• Interactive wardrobe filters, selected fit state, 360 scan completion state, and stylist request state
• Normal/stylist account toggle
• Wardrobe photo-card simulation
• Virtual-you outfit preview
• Style request workflow
• SwitchAI v1 local stylist brain with outfit scoring, explanations, mall suggestions, stylist matching, and feedback learning
• SwitchAI training registry and scripts for real fashion datasets
• Local JSON backend API with persistent profile, wardrobe, scan, role, style request, wishlist, stylist upgrade, social, messaging, mall registration, and competition state
• Real wardrobe, mall, stylist, and social feed images stored in assets/photos/
• Stylist marketplace cards
• Mall wishlist panel
• Social proof feed
• Styling competition card
• Python style-matching engine with tests
• CSV sample style report
• GitHub Pages demo

Run

Full app with backend:

python3 server.py

Open:

http://127.0.0.1:5180

Static frontend only:

python3 -m http.server 5180

Open:

http://127.0.0.1:5180

Test

python3 -m unittest discover -s tests -v

Train SwitchAI

Dataset registry:

python3 tools/dataset_inventory.py --approved-only

Download approved dataset shards:

python3 tools/download_training_datasets.py --dataset fashion_mnist --max-shards 1 --execute

Build the broader approved local corpus with capped downloads:

python3 tools/download_training_datasets.py --max-shards 20 --max-shard-bytes 200000000 --max-total-bytes 700000000 --execute --write reports/training/multi_dataset/download_execute.json
python3 tools/report_training_corpus.py --write reports/training/multi_dataset/corpus_report.json

Train the first vision classifier with managed dependencies:

uv run tools/train_switch_ai_vision.py --dataset zalando-datasets/fashion_mnist --config fashion_mnist --max-train-samples 2000 --max-eval-samples 500 --epochs 1

Train a dependency-free real-data baseline and save metrics/graphs:

python3 tools/train_fashion_mnist_baseline.py

Continue training with a stronger NumPy softmax classifier:

/Users/user/.cache/codex-runtimes/codex-primary-runtime/dependencies/python/bin/python3 tools/train_fashion_mnist_softmax.py

See docs/training.md.

API

• GET /api/health
• GET /api/state
• POST /api/profile/role
• POST /api/profile/measurements
• POST /api/stylist/upgrade
• POST /api/wardrobe
• POST /api/outfit/select
• POST /api/scan
• POST /api/style-requests
• POST /api/ai/style
• POST /api/ai/feedback
• POST /api/wishlist
• POST /api/social/posts
• POST /api/social/react
• POST /api/messages
• POST /api/mall/register
• POST /api/competitions
• POST /api/competitions/entries
• POST /api/stylists/follow
• POST /api/reset

Live Deployment

Live app:

https://switchitup.vercel.app

Deployment files:

• vercel.json routes /api/* to the Python serverless entrypoint in api/index.py
• api/index.py reuses the local backend handler
• app.js reads the optional <meta name="switchitup-api-base"> value and automatically points GitHub Pages to https://switchitup.vercel.app
• server.py uses local JSON by default and switches to Supabase persistence when SUPABASE_URL and SUPABASE_SERVICE_ROLE_KEY are configured
• Each browser gets a persistent session_* id in local storage; the API uses that header to isolate the user state row locally and in Supabase

Supabase setup:

1. Create the table with supabase/schema.sql.
2. Set Vercel environment variables:
   • SUPABASE_URL
   • SUPABASE_SERVICE_ROLE_KEY for a production private backend key; this MVP can also run with the Supabase publishable/anon key plus the single-row RLS policies in supabase/schema.sql
   • optional SWITCHITUP_SUPABASE_TABLE, default switchitup_state
   • optional SWITCHITUP_STATE_ID, default production, used when no valid X-SwitchItUp-Session header is present
3. Redeploy Vercel.

Until Supabase is configured, Vercel uses /tmp JSON state as a live MVP fallback.

Backend Hardening

• Atomic JSON writes with schema migration for newly added state keys
• Supabase/Postgres JSONB state adapter for hosted, per-browser session persistence
• Session-scoped SwitchAI memory for recommendations, feedback, color/category/material preferences, and formality bias
• Request body size limit for upload-heavy JSON requests
• Security headers for local API responses
• Configurable data path, host, port, and CORS origin through environment variables
• Upload validation for PNG, JPEG, and WebP data URLs, including size and file signature checks; local mode writes files, hosted Supabase mode persists validated data URLs in the session state
• Input validation for measurements, stylist plans, mall emails, competition prizes, and social actions
• Unit coverage for SwitchAI recommendations, feedback learning, wardrobe uploads, style requests, wishlist actions, social posts, messages, mall registration, competitions, follows, and measurement updates

Image Credits

Image sources are documented in docs/image-credits.md.

Export Report

PYTHONPATH=. python3 tools/export_sample_report.py

Production Notes

The current app is a functional full-stack MVP with a local JSON backend. A real public production version would replace the JSON store with a managed database, add consent-based body image capture, secure photo storage, product catalog integrations, payments, identity verification for stylists and malls, moderation, privacy controls, and clear AI safety rules.