← All posts

A Spaced Repetition Vocabulary App Without Flashcards

July 3, 2026

Spaced repetition works: review a word right before you'd forget it, and the interval until the next review stretches. That part is settled. What isn't settled is the review itself — and this is where flashcards quietly fail for vocabulary.

Flashcards train recall of a phrasing, not command of a word

A flashcard pairs a word with a definition, and the pairing never changes. Review it twenty times and you have learned exactly that: given this string, produce that string. You start recognizing the definition by its opening words before you've read it. The card gets easier; the word doesn't leap to mind in conversation, because conversation never hands you your definition.

That's overfitting. You memorized the surface form of the card instead of the meaning underneath it — the same failure mode as a model that memorizes its training set. The fix is the same too: vary the surface, keep the underlying target.

What a review looks like in Lexi

I built Lexi so that every review is a fill-in-the-blank sentence plus a hint, delivered inside a crossword generated from the words you're learning:

You retrieve the word from a situation — the way you'll actually need it — with the grid's crossing letters as scaffolding when you half-know it. And the grading happens without you: Lexi classifies your solve speed (fast, normal, slow — normalized for clue and word length) and adjusts the schedule itself. No Again/Hard/Good/Easy buttons, no negotiating with the algorithm after every card.

One word, many surfaces, one schedule

Here is the structural difference, and why you can't rebuild it in Anki. Every word in Lexi cycles through multiple differently-phrased clues — currently three per word:

Three surfaces, one underlying item, one spaced-repetition schedule. There is no fixed phrasing to overfit to, but the scheduler still treats the word as the single thing it is.

Anki can't express this. Put three clue cards for one word in a deck and each card schedules independently — the system now believes you're learning three things, and the word shows up on three interleaved timelines. Collapse them into one card and you're back to a single frozen phrasing. An image-occlusion deck can fake it, but each card carries the full stack of mostly-occluded examples, and you're now the author, editor, and maintainer of thousands of cards.

The clues are the moat

Writing one good cloze sentence per word sounds easy until you try it for a word you're currently learning — you'd need the deep usage knowledge the card is supposed to teach you. Work from the dictionary definition and your sentence quietly encodes your misreading of it.

Lexi's clues come from an iterative pipeline — a creator model drafts, a validator checks it against usage, a guesser has to be able to solve it, a reviser fixes what failed, up to seven rounds per clue before acceptance — with quality criteria that have been refined through repeated revision rounds in the live app. The first ~1,500 words also carry illustrated word cards — and every word you've never seen gets its card shown before the puzzle:

The brutal first flashcard rep — being quizzed cold on something you can't possibly know — doesn't exist here.

Friction is the real enemy

Nobody quits spaced repetition because the math stopped working. They quit because the sessions feel like filing. So the parts of Lexi that look like gamification are doing retention work: notifications fire when words are actually due (not at some fixed hour), streaks and an activity heat map make consistency visible, and there's multiplayer when you want it. Definitions, clues, and images keep improving through in-app feedback and regular updates — a live system, not a static deck.

One more asymmetry: Anki is free on desktop, but the iOS app is $24.99. Lexi is free on iPhone and iPad, and there are no ads.

Who this is for

If you're memorizing anatomy diagrams or kanji stroke order, use Anki — arbitrary self-made decks are exactly what it's for. For building an English vocabulary that you can actually deploy, the flashcard format is the wrong tool, and the mechanism above is the reason. I've learned 1,300+ words with Lexi myself. It's funny, I didn't expect these words to make such a difference but all of a sudden I'm seeing them everywhere, especially in books but even in podcasts, news reports, and even casual conversation. I must have just been ignoring their occurrences when I didn't know what they meant.

Get Lexi on the App Store — free on iPhone and iPad, no ads.