Expert AI/Computer Vision Developer: 100% Local Virtual Try-On Smart Mirror (NVIDIA Edge Hardware)

Project Overview: We are launching a premium retail experience and need an expert developer (or elite agency) to build a fully polished, retail-ready Virtual Try-On (VTO) Smart Mirror. The system will allow customers—specifically focusing on children and youth sizing—to take a photo at a full-sized mirror, input their age/height, select a garment from a local image catalog, and instantly see a high-fidelity, AI-generated image of themselves wearing the item. To maximise generation speed and data privacy, all AI inference must happen locally on-device (e.g., via an NVIDIA Jetson edge computing setup or local GPU PC) with zero cloud-generation dependencies. Additionally, the mirror must feature a "Photobooth" mechanic where users can scan an on-screen QR code to instantly download their generated image to their smartphones. Key Responsibilities: - Hardware Architecture Consulting: Advise on sourcing the optimal full-sized smart mirror hardware and components (via Alibaba or commercial suppliers) paired with a powerful local inference chip (e.g., NVIDIA Jetson AGX Orin or a dedicated local GPU mini-PC setup hidden in the chassis). - Local AI Try-On Pipeline: Adapt and optimise an open-source virtual try-on model (e.g., IDM-VTON, OutfitAnywhere, or similar Diffusion-based pipelines) to run locally and generate results in seconds. - Child-Specific Sizing Logic: Standard VTO models fail on children. You will implement custom logic using age/height inputs to mathematically or visually guide the generation process to reflect accurate child body proportions and garment lengths. - Local Catalog File System: Build an intuitive browsing interface that reads a local directory structure of clothing images and metadata rather than relying on an external e-commerce API. - Local QR Code File Sharing (Photobooth Style): Develop a localised file-sharing system. Upon image generation, the system should generate a QR code linking the user's smartphone to a local micro-web server hosted on the mirror via local Wi-Fi, allowing them to download the photo instantly. - Future-Proofing Architecture: Structure the core AI try-on pipeline cleanly so it can be ported to a consumer web/mobile app in a later phase. Required Technical Expertise: - AI & Computer Vision: Deep experience with Generative AI, Stable Diffusion, ControlNet, and Virtual Try-On (VTON) architectures. - Edge Computing & Hardware: Proven experience optimising AI models for local deployment using NVIDIA TensorRT, CUDA, and NVIDIA Jetson or Windows/Linux edge-GPU hardware. - Sizing & Proportions: Experience with pose estimation (OpenPose/MediaPipe) and image wrapping/warping to handle precise body dimensions (specifically pediatric/child proportions). - Local Networking: Experience setting up local web servers (Node.js, Flask, Python, etc.) and QR code delivery systems for offline/local network file transfers. - UI/UX Development: Experience building sleek, Kiosk-mode interactive applications (React, Electron, Python/Qt, etc.) for vertical touchscreen formats. Project Timeline & Deliverables: Phase 1 (Weeks 1-2): Hardware selection/ordering and architecture blueprint. Phase 2 (Weeks 3-5): Local AI VTON pipeline optimisation and child-sizing logic implementation. Phase 3 (Weeks 6-7): Local database UI, asset pipeline, and QR/Photobooth networking layer. Phase 4 (Week 8): Testing, calibration, UI polishing, and deployment preparation. Hard Deadline: Fully functional retail deployment by September 1, 2026. To Apply, Please Provide: - Examples of edge-AI or local GPU-inference hardware projects you have built. - Any experience you have with Virtual Try-On technologies or image-to-image garment diffusion.