Lookitry

AI-Powered Virtual Try-On for B2B E-Commerce in Latin America

Empower fashion brands to integrate a virtual try-on widget into their stores in minutes, reducing return rates and boosting conversions. "Try it before you buy it."

---

Pitch & Value Proposition

Lookitry is a B2B SaaS platform designed to revolutionize the way fashion, accessories, and footwear are purchased online in Latin America. Powered by advanced artificial intelligence, final customers can upload a selfie and instantly visualize how a specific garment fits them directly from the brand's e-commerce store.

Our solution integrates into stores in two primary ways:

• Widget Script (/widget.js): A lightweight, highly optimized script that embeds the virtual try-on experience as a floating or inline button on any major e-commerce platform (Shopify, WooCommerce, Wix, or custom developments).
• Customizable Mini-Landings: High-converting landing pages designed with premium aesthetics where brands can direct social media traffic to offer friction-free virtual try-on sessions.

---

System Architecture

Lookitry's architecture is engineered for high availability, robust security, and a strict separation of concerns:

graph TD;
    A["Client / E-commerce"] -->|"Widget Script / Mini-Landing"| B("Frontend: Next.js 14 App Router")
    B -->|"REST API via Dual JWT"| C("Backend: Express Node.js 22")
    C <-->|"PostgreSQL / Service Role"| D[("Supabase")]
    C -->|"Upload / Storage"| E[("MinIO S3 Compatible")]
    C -->|"Queue / Cache"| F("Redis Cache & Job Queue")
    C -->|"FastAPI / Local CPU"| G("MobileSAM Service")
    C -->|"Secure Webhook"| H("n8n Orchestrator")
    H <-->|"API"| I("Google Vertex AI / Gemini 2.5")
    H -->|"Generated Outputs"| E
    C <-->|"COP Payments"| J("Wompi")
    C <-->|"USD Payments"| K("PayPal")

Loading

Virtual Try-On Generation Pipeline

1. Request: The user uploads a selfie and selects a garment from the Frontend.
2. Validation: The Backend validates the session, confirms available brand credits, and enqueues the job in Redis.
3. Segmentation: A dedicated local MobileSAM microservice (Python/FastAPI running on CPU) segments the human silhouette and generates a precise PNG mask in seconds.
4. Inpainting & AI: The segmented mask and garment image are dispatched to the primary AI pipeline in Google Vertex AI, using Gemini 2.5 Flash Image to execute highly realistic garment inpainting.
5. Delivery: The generated try-on output is optimized and saved in MinIO's S3-compatible storage, updating Supabase for real-time polling from the client's screen.

---

Tech Stack

Frontend

• Framework: Next.js 14.2.35 (App Router)
• Language: TypeScript 5.9.3
• Styling: Tailwind CSS 3.4.0 (Dark-by-default theme with pristine contrast and fluid typography)
• Motion & Micro-interactions: Framer Motion & GSAP for high-end cinematic web experiences
• Image Optimization: Sharp for on-the-fly compression and ultra-fast WebP delivery

Backend

• Runtime: Node.js 22 + Express (Highly modular, clean, and robust architecture)
• Security:
  • Dual-JWT authentication with key rotation (Access Token + Refresh Token stored in HTTP-only cookies)
  • Distributed Redis-backed Rate Limiting to prevent API abuse
  • Form protection and anti-spam supported by Cloudflare Turnstile
• Cache & Throttling: Redis (ioredis) for caching metadata and scheduling background task queues

Database & Storage

• Database: Supabase (PostgreSQL) leveraging pgvector for RAG vector similarity search
• Storage: MinIO (S3-compatible) for local and federated assets, protecting client data privacy

---

Core Engineering Highlights

• Dynamic Currency Conversion: An elite, secure system to automatically convert COP plans to USD based on the real-time TRM plus a dynamic safety margin, protecting SaaS unit economics.
• RAG & Knowledge Base (Rebecca): A fully automated sales and customer-service assistant (available on web and WhatsApp) that leverages pgvector and Gemini Embedding 001 to resolve brand FAQs, feature details, and onboarding queries directly from official documentation.
• Microservice Container Deploys: Dockerized service isolation (Backend, Frontend, MobileSAM, n8n, Redis) behind Traefik as the main reverse proxy, permitting surgical zero-downtime updates.
• API Widget Protection: A security middleware checking authorized origins (allowed_origins) combined with 1-hour Redis configuration caching to prevent script hijacking and hotlinking.

---

Repository Structure

LOOKITRY/
├── frontend/                    # Next.js 14 App Router - Dashboards, checkout, landings
├── backend/                     # Express API (Node 22) - Core business logic, payments, security
├── sam-service/                 # Python/FastAPI MobileSAM - On-premise human segmentation
├── lookitry-woocommerce/       # Native PHP WordPress plugin for custom WooCommerce setups
├── mcp-gcp/                     # GCP MCP Server for cloud storage & compute automation tasks
└── error-pages/                 # Maintenance & graceful error landing pages

---

Quick Start (Development)

Prerequisites

• Node.js >= 20
• pnpm == 9.15.9 (Mandatory to avoid npm package vulnerability exploits)
• Docker & Docker Compose

Local Setup

1. Clone the repository:

   git clone https://github.com/depper-IA/Lookitry.git
   cd Lookitry

2. Configure environment variables:

   • Copy frontend/.env.example to frontend/.env
   • Copy backend/.env.example to backend/.env

3. Install packages via pnpm:

   pnpm install

4. Spin up local development containers:

   docker compose -f docker-compose.dev.yml up -d

---

Designed and built with passion to lead the future of visual commerce.
© Lookitry. All rights reserved.