How a lesson is made
A lesson is a video's transcript turned into something you can read along with: every word clickable, with its reading, its meaning in that sentence, and an English line under each sentence.
1. The transcript
YouTube videos often have several caption tracks. The pipeline reads them all and picks the one closest to what's said: the creator's own captions first, then YouTube's automatic captions. For Cantonese this matters more than for most languages, because many videos are subtitled in written Chinese, which follows the meaning rather than the words spoken. Those subtitles are kept as a second line you can turn on, but the lesson follows the speech.
The language is detected from the caption tracks and the language the uploader set, and where those are unclear (Chinese labels cover both Cantonese and Mandarin), from a sample of the speech, so you only paste a link.
2. Words and readings
Each line is split into words, and each word gets a reading (Jyutping or Yale for Cantonese, pinyin for Mandarin, furigana over the kanji for Japanese) and the senses listed in open dictionaries (see Dictionary). Where two tools disagree on a reading, the word is flagged for the next step.
3. English, meanings and notes
English comes from the creator's own English captions when the video has them. Otherwise it's written by an AI model, which also picks the right meaning for each word in its sentence, fixes readings and word boundaries, and adds short notes on particles and slang. It can't change the transcript: anything that looks misheard is listed for a person to check instead.
4. Timing
Lines follow the video using the caption timings. When the captions time each word, as YouTube's automatic captions do, the underline steps from word to word as they're spoken; otherwise it moves through the line at an even pace.
Levels
Beginner, Intermediate and Advanced come mostly from how fast people speak in the video, since that's what makes listening hard, with a step up for videos that use many uncommon words.