CHANWOO
SONG 한국어 ↗
← All work
CONTENT PIPELINE / OFFLINE2026.09

Tago
Voca

Follow unfamiliar words without losing the context.

INDEPENDENT PRODUCT / RELEASED

Case study by

Released app · word detail and in-context explorationReleased app · word detail and in-context exploration
Released app · word detail and in-context exploration
01 / CONTEXT & CONTRIBUTION

The problem.
My part in it.

Background

  • External lookups interrupted exploration of unfamiliar words.
  • Definitions, examples, audio and review needed a continuous flow.
  • Runtime generation would add waiting time and recurring cost.
  • The app needed validated content available without a network.

Objective

  • Explore word → example → another word, up to 20 levels.
  • Support local exploration, listening and review with verified content.

My role

  • Independent product / content architecture and AI-assisted build.
  • Owned word links, data/audio validation, packaging and release.
02 / SYSTEM DESIGN

How it connects.

Production & device boundary
Content production and local learning are separate systemsGenerate and validate before distribution; explore and review on the deviceContent production and local learning are separate systemsGenerate and validate before distribution; explore and review on the deviceCONTENT PRODUCTION & DISTRIBUTIONON-DEVICE LEARNINGPackaged releaseSource vocabulary — 10,635 vocabulary itemsSourcevocabulary10,635 vocabulary itemsGenerate & review — Examples / translations / imagesGenerate &reviewExamples / translations/ imagesAudio & validation — Pre-generation / hash checksAudio &validationPre-generation / hashchecksContent release — SQLite package + audioContent releaseSQLite package + audioSQLite content DB — Read-only content / Validated audioSQLite contentDBRead-only contentValidated audioExplore · listen · review — Word → example → another wordExplore · listen · reviewWord → example → another wordLearning record DB — Personal state and history / Stored locallyLearning record DBPersonal state and historyStored locallyOptional account sync; core learning does not require generation on each request.
Zoom for detail
Processing / data flowIteration / feedbackService / environment boundary
01

Generate before the session.

Prepare and validate content before distribution. Keep word exploration, listening and review available locally without generating every item on demand.

02

Separate content from progress.

Read-only content and audio have their own validation and version lifecycle. Store personal progress and learning history in a separate local database.

03 / RESULTS & SCOPE

The outcome.

12,464Learning entries
10,635Source vocabulary items
20Exploration levels

September 2026 content records. Entry counts describe the packaged learning content.

NEXT PROJECT / 07Agent Memory↗