Lime 🍋 — AI-Powered Digital Wardrobe

Lime turns a phone photo of a clothing item into a tagged, searchable entry in your digital closet — then uses vector search and live weather data to recommend outfits that actually go together.

Live app: lime-wardrobe.vercel.app AI service: andrew-heejay-lime-backend.hf.space

  

Screenshots above are from the public sign-in flow. Wardrobe, ingestion, and the swipe deck require an account — see Getting Started to run it locally.

What it does

• Vision ingestion pipeline — snap a photo of one item; a FastAPI service removes the background (briaai/RMBG-1.4), tags it with Gemini 2.0 Flash (category, silhouette, palette, texture, aesthetic), and embeds it for similarity search. No manual data entry.
• Weather-aware recommendations — on the styling deck, Lime fetches your local weather (Open-Meteo), turns it into a natural-language "ideal outfit" description, and re-ranks your wardrobe by aesthetic fit using vector search (Pinecone).
• Tinder-style swipe deck — swipe through tops, bottoms, and shoes (framer-motion) to put together outfits.
• Wardrobe grid — browse and filter your full digital closet by category.

Tech stack

Layer	Technology
Frontend	Next.js 16 (App Router), React 19, TypeScript, Tailwind CSS v4, shadcn/ui, framer-motion
Auth, database & storage	Supabase (Postgres, Auth, Storage) with Row Level Security
AI service	FastAPI (Python)
Background removal	briaai/RMBG-1.4 via 🤗 transformers
Vision tagging	Gemini 2.0 Flash, structured output validated with Pydantic
Embeddings & search	sentence-transformers/all-MiniLM-L6-v2 → Pinecone
Weather	Open-Meteo
Hosting	Vercel (frontend) · Hugging Face Spaces / Docker (AI service)

Architecture

Two independently deployed services joined by a shared UUID: Next.js owns UI, auth, and orchestration; FastAPI owns the entire ML pipeline and stays stateless.

flowchart LR
    U[Browser] -->|photo| NJ[Next.js / Vercel]
    NJ -->|POST /items/ingest| AI[FastAPI / HF Spaces]
    AI --> BG[RMBG-1.4] --> VLM[Gemini 2.0 Flash] --> EMB[MiniLM embedding]
    EMB --> PC[(Pinecone)]
    AI -->|tags + image| NJ
    NJ -->|store| SB[(Supabase)]

    U -->|geolocation| NJ
    NJ -->|GET /recommendations| AI2[FastAPI]
    AI2 --> OM[Open-Meteo] --> PC
    PC -->|ranked items| NJ

Loading

Full write-up — including the ingestion and recommendation flows, data model, and key design decisions — is in docs/ARCHITECTURE.md.

Getting started

Prerequisites

• Node.js 20+ and Python 3.11+
• A Supabase project with supabase/schema.sql applied
• A Pinecone index (384 dimensions, cosine metric)
• A Gemini API key

Backend (FastAPI)

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env   # fill in GEMINI_API_KEY, PINECONE_API_KEY, PINECONE_INDEX_NAME
uvicorn app.main:app --reload --port 8001

Frontend (Next.js)

cd frontend
npm install
cp .env.local.example .env.local   # fill in Supabase URL/anon key + BACKEND_URL
npm run dev

Visit http://localhost:3000 — you'll land on /login and can sign up for a new account.

Project structure

.
├── frontend/    # Next.js app — UI, auth, server actions
├── backend/     # FastAPI AI microservice — tagging, embeddings, recommendations
├── supabase/    # Postgres schema + RLS policies
└── docs/        # Architecture, product requirements, technical design

Documentation

• docs/ARCHITECTURE.md — system design, pipelines, data model, key decisions
• docs/PRD.md — product requirements and user journey
• docs/TECH_DESIGN.md — original technical design and cost analysis

License

MIT