Philosophy¶
The best learning method is no fixed method.
That's the thesis at the heart of AdaptiveLearner. Most "learning apps" pick one approach — flashcards, video lectures, gamified streaks — and assume everyone learns the same way. They don't.
Why one method isn't enough¶
Different topics call for different methods. Even the same topic, at different stages of mastery, calls for different methods:
- You don't approach learning a new grammar rule the same way you approach polishing your accent.
- You don't review for a high-stakes exam the same way you explore a topic out of curiosity.
- You don't recover from a setback the same way you maintain a streak.
A learner who can fluidly switch between methods learns faster, retains longer, and burns out less. The point of AdaptiveLearner is not to find your One True Method — it's to make method-switching cheap, natural, and pedagogically justified.
The six methods¶
We picked six methods that cover the major axes of how humans actually learn:
| Method | Core stance |
|---|---|
| Deductive | Theory first — rules, then examples |
| Inductive | Examples first — derive rules from patterns |
| Error-based | Provoke mistakes, learn from them |
| Dialogic | Conversational exchange, low pressure |
| Contextual | Real-world scenarios, situated practice |
| AI-adaptive | Let the AI choose per turn |
The first five are pedagogically classical. The sixth is the honest acknowledgment that AI can do something humans couldn't: pick a method per individual exchange based on what the learner just did.
The seven-step cycle¶
Each session walks through a 7-step learning cycle: Input, Attempt, Error, Feedback, Adapt, Repeat, Integrate. Most learning happens between Error and Feedback (steps 3-4) — that's where the actual cognitive work lives. The other steps exist to give errors something to push against and a place to land.
The cycle isn't a conveyor belt. Steps repeat, skip, or step back depending on whether you actually grasped the material. The dual-prompt AI architecture (see The seven-step cycle) decides per round- trip.
Git for learning¶
We borrow Git's mental model for tracking progress:
- Commit = one session's snapshot (method, ratings, duration).
- Diff = the delta from the last session in the same topic.
- Branch = a method-switch (you went from deductive to dialogic at session 7).
- History = your full learning trail, queryable + visualisable.
We don't use Git literally; we use its discipline of versioned, recoverable, comparable state. ChatGPT forgets your conversation the moment you close the tab. AdaptiveLearner keeps a structured snapshot of every session so trends emerge across weeks and months.
The three pillars¶
Three external tool categories sit alongside AdaptiveLearner sessions — we don't try to reinvent them:
- Spaced repetition (Anki) — for long-term retention of rules + error-corrections.
- Active recall (NotebookLM) — for building knowledge from your own sources.
- Adaptive AI prompts (Claude / ChatGPT / Gemini) — for one-off explanations + flexible inquiry.
The Dashboard's Tool Recommendations card ranks these five tools (plus Excalidraw + Obsidian) against your profile, so you see which one to lean on for your current learning shape.
Why this matters¶
The Medium article series Von Theorie zur Praxis is the intellectual foundation. AdaptiveLearner is the engineering translation: a tool that operationalises the article's arguments into a daily practice. The articles explain WHY six methods, WHY seven steps, WHY adaptive switching. This app gives you a place to actually do the thing.
The articles aren't required reading to use the app. But if you want to understand why we built it this way, that's the source. Links live in the README.