LiveConsumer Apps
HangeulSprint
Korean drills with audio, no romanisation crutch.
The Problem
Romanisation trains the wrong reflex. HangeulSprint is drill-heavy: listen, repeat, short sessions, closer to ear training than streaks and badges.
Architecture
Next.js 15 with SST deployment
API Gateway + Lambda backend
S3-hosted audio assets
Cognito authentication
CDK infrastructure stack
Engineering Decisions
Why speechSynthesis for some lines?
Avoids Polly billing for every sentence. Browser speaks Korean when the voice exists; fixed drills use files.
Where do drills come from?
Generated from vocabulary/grammar data so adding a layer is mostly data entry, not writing hundreds of questions by hand.
Highlights
Audio-first drillsLayers get harder as you goProgress stored per sessionUses browser speechSynthesis for ad-hoc lines; S3 clips for fixed drills
Stack
Next.js 15Tailwind v4SSTCDKAPI GatewayLambdaS3Cognito