Building an Offline Face Check-In with LiteRT.js
Google's LiteRT.js launch post promised web AI inference with maximum performance, entirely locally. I wanted to see if that held up on something real, not a benchmark demo, so I built a small proof of concept: face check-in that runs fully offline in the browser, with a total model footprint under 5 MB. Here's the pipeline, the license detour I almost walked into, and a liveness bug that would have made the whole thing unusable.