What I learned building an AI-driven spaced repetition app
a day ago
- #spaced-repetition
- #AI
- #learning
- The author built an AI-driven spaced repetition app called AutoDeck, which generates infinite flashcards on any topic and adjusts difficulty based on user responses.
- Spaced repetition is an effective learning method where material is reviewed at increasing intervals, helping to internalize knowledge over time.
- Traditional spaced repetition tools like Anki require users to create their own flashcards or find pre-made decks, which can be time-consuming and may not match the user's learning needs.
- AutoDeck leverages LLMs to eliminate the setup cost of creating flashcards, allowing users to start learning immediately by simply specifying a topic.
- The app avoids a chat interface, which is common in AI apps, to reduce user effort and improve usability.
- Generating an infinite feed of flashcards with AI presented challenges related to speed and consistency, with solutions involving batching and streaming card generation.
- The author used OpenAI’s Codex to build most of the app, noting that while it saved time, it still required significant oversight and intervention to correct mistakes.
- Building AI projects often necessitates charging users due to the costs associated with AI inference, which the author sees as a positive forcing function for quality.
- The author found that GPT-5 was not ideal for generating flashcards due to slow response times and awkwardly concise outputs, suggesting that less advanced models may suffice for this use case.