Public learning pages
Landing, about, grammar case, and preview routes introduce structured language learning content.
Case study
Slovka
A multilingual vocabulary platform combining public learning content, protected study workflows, progress tracking, cultural-text ingestion, image generation, and AI-assisted vocabulary pipelines.
01 / 05
Product inventory
Repository evidence: routes, data model, interface screens, and workflow code.
- Page routes
- 22
- Drizzle tables
- 40
- UI components
- 302
- Pipeline files
- 795
- Locales
- 4
Project overview
A multilingual vocabulary platform combining public learning content, protected study workflows, progress tracking, cultural-text ingestion, image generation, and AI-assisted vocabulary pipelines.
Ukrainian vocabulary learning does not fit a flat flashcard model. Collections, nested lists, words, translations, grammar references, images, quizzes, progress, goals, reading lessons, generation jobs, and review workflows all need to stay connected without making the learner-facing pages feel like an admin tool.
What was built
A language-learning system where cultural-source ingestion, linguistic normalization, vocabulary modeling, learner progress, media generation, multilingual UX, and AI-assisted review workflows stay connected.
Learner workspace
Protected collections, lists, words, quizzes, progress, goals, calendars, and kids reading pages.
Content operations
Admin and tool routes manage collection generation, song import, lyric and poem processing, image regeneration, review, reconciliation, and deduplication.
Pipeline infrastructure
Graph-based pipelines crawl carols and poems, extract dictionary-form vocabulary, translate, reconcile existing words, queue imagery, and persist collections; one carol run processed several dozen songs.
Pipelines
Artifacts, gates, telemetry, reconciliation.
Media storage
R2-backed image workflows.
Selected details
Implementation details from the product, data, workflow, and infrastructure layers.
- 01
Vocabulary is modeled through collections, lists, words, translations, documents, and membership tables.
- 02
Learning state includes quizzes, statuses, study plans, goals, and calendar data.
- 03
Song pipelines crawl carol sources, preserve lyrics and metadata, and turn each song into a bilingual vocabulary list.
- 04
Lyric and poem pipelines chunk longer texts, extract learner-facing vocabulary, standardize entries into dictionary forms, deduplicate candidates, and reconcile against existing words.
- 05
One Christmas carol run processed 19 songs and 1,612 source words; a poem-length run consolidated 328 raw vocabulary candidates into 288 unique entries before persistence.
- 06
Image-generation workflows create consistent vocabulary imagery, upload assets, and attach generated documents back to accepted word records.
- 07
AI-generated content is routed through artifacts, schemas, review gates, reconciliation, image queues, and persistence steps.
- 08
Cloudflare deployment uses Workers, D1, R2, and OpenNext-oriented application structure.
- 09
Learner-facing routes and internal tools stay separated while sharing the same vocabulary, image, and collection records.
Systems overview
How the public pages, staff workflows, data, review steps, and infrastructure connect.
Learning pages
Landing, grammar, previews.
Content graph
Collections, lists, words, translations.
Learner tools
Quizzes, progress, goals, kids reading.
Operations
Admin, import, generation, deduplication, review.
Pipeline layer
Crawling, extraction, image queues, persistence.
Technology
The platform stack behind the build.
Design language
A tactile Ukrainian study system: literary typography, floral framing, pale peach fields, sage and coral accents, circular word imagery, and denser review tools for content operations.
Philosophy
Study pages with operational depth
The public learning pages stay warm and tactile while the same content model supports protected progress, review queues, image generation, and pipeline persistence.
Floral study framesCircular word mediaTabbed grammar panelsReview-gated generation
Aa
Serif display, text, Fira CodeStudy pages, vocabulary, pipeline state
Ink#28332d
Paper#fff7ef
Sage#9caf88
Coral#db715c
Blush#f4d6c8
Selected screens
Additional screens from the case study.