Front-end dev here. Over the last few months I shipped <a href="http://getfacereport.com/">Get Face Report</a> - a tool that analyzes facial proportions entirely in the browser, no uploads. Sharing what I learned, since camera + on-device processing comes up in a lot of front-end projects and interviews.
1. getUserMedia UX is 80% of the work.
The API itself is simple:
<pre class="shiki github-dark shaku"><code>const stream = await navigator.mediaDevices.getUserMedia({ video: { facingMode: 'user', width: 1280 }});</code></pre>
The hard part is the failure paths: permission denied, no camera at all, insecure origin (getUserMedia requires HTTPS), and users who denied the prompt and expect a way to re-enable. Surface a specific error state for each case - a generic "camera error" message kills conversions.
2. On-device processing is a privacy feature you can actually market.
Everything - landmark detection, geometry measurements - runs locally on a canvas. No frames ever leave the device. That simplifies compliance and makes the privacy story trivial to explain: your photo never leaves your browser.
3. Performance tips that mattered:
<ul>
<li>Downscale the video frame before analysis; you rarely need full 1080p for landmark detection.</li>
<li>Throttle the processing loop to around 10fps - users cannot tell the difference.</li>
<li>Use <code>requestVideoFrameCallback</code> where available instead of rAF to get frames actually synced to the camera.</li>
</ul>
4. Test on a mid-range Android.
Front cameras differ wildly in exposure and white balance, which affects detection quality. The worst failures never showed up on my laptop.
If you are curious about the result, the free scan lives at <a href="http://getfacereport.com/">Get Face Report</a> - it maps face shape, symmetry and proportions and turns them into a personal styling plan. Feedback from front-end folks especially welcome.
What has your experience been with getUserMedia projects? Any tricks for the permission re-prompt flow?

Lessons from building an in-browser face analysis tool (getUserMedia + on-device inference)

It's a good idea to prepare a short self-introduction, not only for interview but also for other social purposes.…