Simple request: I have the Olly Richards Short Stories in Russian audiobook (25 MP3 chapters) and the matching EPUB. Can you make Anki cards from it?

Here’s what “yes” actually involved:

Step 1: Figure out what kind of cards to make. Not obvious. Vocabulary cards? Sentence cloze? Listening comprehension? I/O format mattered a lot for SRS effectiveness. Claude spun up four simultaneous research agents — polyglot community strategies, popular Anki addon consensus, Reddit recommendations, published spaced repetition science — and synthesized a deck architecture: vocabulary with audio, cloze deletions, listening comprehension (audio → meaning), and grammar glosses for construction patterns.

Step 2: Get the text. The EPUB had Unicode stress marks (acute accents on Russian vowels) scattered through the text. The parser needed to handle those without stripping them, because they’re pedagogically important for a learner. Span joins across page boundaries were also an issue — sentences that crossed EPUB sections needed to be reassembled.

Step 3: Get the audio aligned to sentences. Whisper transcribed all 25 chapters to SRT format, then the segments were aligned to the EPUB text so each card could have the corresponding audio clip.

Step 4: Ordering. Raw frequency ordering isn’t ideal for learners — you want i+1, roughly: words you almost know, not words you’ve never seen. FrequencyMan handles this as an Anki addon. The deck was structured so FrequencyMan could reorder at review time.

Step 5: Push to Anki. AnkiConnect API, running locally. 4,198 cards.

The session ended with me asking where the cards were — I couldn’t find them. Claude confirmed they were on my Mac. I had assumed the ThinkPad. After 646 lines of session logs, the resolution was: check the other machine.

That’s a pretty honest picture of what AI-assisted language learning infrastructure looks like at the seams. The hard parts are all data wrangling and API plumbing. The easy part is the card generation itself. And the very last moment — “wait, which computer?” — is still completely human.