Magic Mirror AI / AR Platform
Interactive AI/AR kiosk experience for retail and event activations
3 months to first deployment
- Detection pipeline
- Real-time, tuned per-device
- Frame rate
- Stable under kiosk constraints
Overview
An interactive kiosk product combining real-time computer vision with AR overlays, built for retail and event brand activations that needed to stand out with a hands-on, camera-driven experience rather than a static display.
The Problem
Retail and event clients wanted an interactive brand activation that felt novel and camera-driven, not another static screen — but kiosk-grade hardware and unpredictable venue lighting made real-time computer vision genuinely difficult to keep stable.
Architecture
A React-based kiosk frontend captures camera input on-device, runs it through a face/pose detection pipeline, and renders AR overlays in real time. A backend service handles session management and media capture/sharing so attendees can take their result with them.
Challenges and solutions
Challenges
- Maintaining real-time performance on fixed, non-upgradable kiosk hardware
- Handling ambient lighting variability across different retail and event venues without degrading detection accuracy
Solutions
- Tuned the detection pipeline and frame throttling specifically against the target kiosk hardware profile, not a development machine
- Added a lighting calibration step at session start to normalize detection input across venues
Performance Improvements
- Achieved stable real-time overlay rendering at interactive frame rates on constrained kiosk hardware
Business Impact
Delivered a differentiated, camera-driven brand activation for retail and event clients, functioning reliably across varied venue conditions rather than only in ideal demo conditions.
Lessons Learned
Interactive AI/AR products have to be designed for the worst-case hardware and lighting the product will ever see, not the best-case demo environment.
Stack
- React
- Node.js
- AWS
- Computer Vision