Introducing Elephant: A Korean Workbook That Talks Back
Why I built a study companion that gives you feedback on your sentences — and what makes it different from flashcard apps.
If you’ve studied Korean for any length of time, you’ve probably built a comfortable Anki routine. You review your cards, you grind through the deck, and the retention numbers look great. But then you sit down to write a sentence — and nothing comes out right.
That gap between recognising vocabulary and using it was what pushed me to build Elephant.
What Elephant Is
Elephant is a Korean study workbook. You write sentences, attempt translations, and read numbers aloud — and you get instant AI feedback on what you got right, what you got wrong, and why.
It’s not trying to replace your Anki deck. It’s not a course. It’s the thing you use alongside those tools: the place where you put your vocabulary under pressure and find out how much of it actually stuck.
Why FSRS?
Elephant uses FSRS, the scheduling algorithm developed by Jarrett Ye and the open-spaced-repetition community — the same algorithm that powers modern Anki.
FSRS models memory with two components: stability (how long a memory lasts) and difficulty (how hard an item is for you specifically). Compared to older SM-2-based algorithms, it tends to schedule fewer reviews while maintaining the same retention target. Less grinding, same results.
When you practice vocabulary in Elephant, the app tracks each word individually and uses FSRS to decide when you should see it again. You’re not just writing sentences — you’re building a personalised review schedule based on real usage.
What’s Next
The core workbook features are live. Over the coming weeks I’ll be shipping:
- Flashcard sessions — a dedicated FSRS review mode for vocabulary
- Vocabulary overview — browse everything you’ve studied, filter by status, see your retention curve
- Translation tasks — structured practice with AI-evaluated feedback on grammar, nuance, and word choice
If you have feedback, ideas, or run into a bug, open an issue on the GitHub repo or reach out directly.
Thanks for trying it.
— David