need help for an offline-first CRDT sync architecture

[amirhosseinadev]

Hello everyone,

I'm in the design phase for a sync feature in an offline-first flashcard app I'm building with Next.js and json-joy. The app needs to support manual, multi-device sync via cloud storage (like Google Drive) and must be performant as a user's dataset grows to 100,000+ flashcards. My initial thought was to use a single, large CRDT Model for the entire user dataset, but my early benchmarks have raised significant performance concerns with this approach.

I'm now re-evaluating and would be very grateful for some guidance and ideas for this scenario. my main question is what your recommendations are to handle large datasets and keep them conflict free.

my data structure currently looks like this. this is my dexie.js input:

{
			"words":
				"id, type, createdAt, updatedAt, dictionaryUrl, vocabularyLevel, word, [word+type], pronunciation, translation, definition, examples, isDeclinable, forms, isComparable, category, wordOrderEffect, gender, elements, numericValue, government, auxiliaryVerb, separability, transitivity",
			"fsrs_cards":
				"id, wordId, createdAt, updatedAt, stability, difficulty, retrievability, due, state, reviews, lapses, scheduled_days, reps, last_review",
			"fsrs_reviews":
				"id, cardId, createdAt, updatedAt, rating, duration, scheduled_days, review_date, stability_after, difficulty_after, state_after",
			"notification_settings":
				"id, createdAt, updatedAt, enabled, frequency, morningTime, eveningTime, customIntervalHours, streakRemindersEnabled, streakReminderDelayHours, minimumCardsDue, permissionGranted, lastNotificationSent, pushSubscription",
		}