Memdora: Designing Cognitively-Grounded Flashcard Interactions for AI-Powered Spaced Repetition

Synthesis: Presents Memdora, a cross-platform AI spaced repetition system that addresses limitations of binary flip-and-rate flashcard interactions. Grounded in cognitive science evidence on retrieval practice, it enables richer interaction patterns and reduces context-switching by generating flashcards from reading material. Demonstrates improved retention compared to traditional SRS tools.

Key Findings

• Spaced repetition systems have robust effects on long-term retention, but existing tools reduce flashcard interaction to a single binary gesture: flip and self-rate, an impoverished model that fails to leverage decades of cognitive science evidence on retrieval practice.
• Memdora contributes a taxonomy of 17 cognitively-grounded interaction types across three learning categories — Language (6 types), By Heart (1 type with 3 retrieval modes), and Exam (10 types) — each mapped to peer-reviewed cognitive science evidence displayed on every card.
• A unified AI generation pipeline collapses card creation to a single gesture at the point of reading, across web, mobile, and three browser extensions (Chrome, Edge, Firefox), reducing the need to context-switch out of reading flow.
• A collaborative classroom layer enables teachers to publish decks, assign them to students, and track learning outcomes at the individual card level.
• An effort-based behavioral reward system incentivizes actual cognitive engagement rather than mere app presence, and the system integrates FSRS-6, the current state-of-the-art spaced repetition algorithm.
• Memdora is deployed publicly on iOS, Android, Web, and three browser extensions, advancing beyond prior AI flashcard systems including SmartFlash and KARL.

Design Rationale

The design rationale treats the flashcard as a full retrieval-practice instrument rather than a passive review object. The forgetting curve first described by Ebbinghaus — roughly 70% of newly learned material forgotten within 24 hours without review — motivates scheduling, while the interaction taxonomy maps each interaction type to peer-reviewed evidence, so learners see the cognitive rationale behind the activity. The effort-based reward system extends this by rewarding the actual work of retrieval rather than time spent in the app, connecting to Self-Regulated Learning and to retrieval-practice research.

What this means for practice

• Learners. Replace flip-and-rate review with a retrieval type matched to the material: Memdora's 17 interaction types (6 Language, 1 By Heart with 3 retrieval modes, 10 Exam) each map to cited cognitive-science evidence.
• Learners. Generate cards at the point of reading instead of in a separate session — the pipeline collapses creation to a single gesture across web, mobile, and three browser extensions, avoiding a context switch out of reading flow.
• Instructors. Publish and assign decks, then use per-card tracking to find the cards students are failing, since the classroom layer reports outcomes at the individual card level.
• Software developers. Reward retrieval effort rather than app presence and schedule with a current algorithm: Memdora couples an effort-based reward system with FSRS-6 and a default 90% target retention.
• Software developers. Instrument the transparency hypothesis — that displaying the cognitive-science citation on every card raises learner trust and motivation — rather than assuming it, since the authors treat it as a testable design hypothesis.

Limitations

• No controlled user study: the efficacy of the 17 interaction types within Memdora, alone or in combination with FSRS-6 scheduling, has not been empirically evaluated, and a longitudinal comparison of retention across card types is still planned.
• The classroom features (teacher assignment, per-card outcome tracking, institutional authentication) have not been evaluated in a real educational setting, so unmet teacher-workflow and class-management needs may remain.
• AI-generated cards may vary in quality in highly specialized domains where the underlying language model has limited training coverage; user editing and one-click regeneration mitigate the variance but do not remove it.
• The retention and reward claims are design rationale: the effort-reward comparison on 30-day retention and the classroom evaluation against control groups using traditional study methods are planned, not reported.

Citation

Ruiyang Zhang (2026). Memdora: Designing Cognitively-Grounded Flashcard Interactions for AI-Powered Spaced Repetition.