Web & SystemsIN DEVELOPMENT· Live and usable. Features are still being improved.

FlazhAI

AI study tool that turns a YouTube video or a document into a spaced-repetition flashcard deck

Timeline

2025 → 2026 (nights and weekends)

Role

Solo. Design, engineering, infrastructure, and copy

Team

1 person

Visit FlazhAI ↗

OVERVIEW

FlazhAI is an AI study tool that turns a YouTube video or a document into a ready-to-review flashcard deck, then tells you what to review each day with spaced repetition. It is live and usable, and I'm still improving its features.

CONTEXT

Most flashcard tools make you write the cards yourself, which is the part nobody does. The idea behind FlazhAI is that the deck should already exist by the time you decide to study.

THE CHALLENGE

Speed: cards should appear while the model is still writing them, not after.
Reliability: YouTube blocks anonymous transcript fetching from server IPs, so a missing transcript must not break generation.
Quality: duplicate or near-duplicate cards should never reach the deck.

THE SOLUTION

Streamed card-by-card generation with the Vercel AI SDK; each card is saved to Postgres (Supabase) as it is parsed and the client receives it over server-sent events.
A three-layer transcript fallback (database cache, then proxy-based fetch, then a backup provider) so failures degrade gracefully.
Inline deduplication during the stream plus near-duplicate detection with pgvector embeddings.
Spaced repetition on a tuned SM-2 scheduler with smart daily review queues.
English and Indonesian localisation.