Level 1
Primary school
Grades 1 to 4. Very short sentences, everyday examples instead of technical terms. Age-appropriate language (ages 6 to 10). No LaTeX.
AI flashcard generator at your level
Enter a topic and your context. Quanta creates cards with selectable Bloom levels 1 to 6, checks them against existing cards and prepares the set for FSRS review.
European AI provider (Mistral, Paris) · source-first default · unverified opt-in
Start learning now
Starter is free · one AI run included
„Der größte Vor[t]eil […] ist, dass die KI jede Karte mit Hilfe seriöser Quellen erstellt und diese Quellen auch als Link anzeigt, sodass man dort selber nachlesen kann."- TU Dresden
- Humboldt-Universität zu Berlin
- HTW Dresden
- Fachhochschule Dresden
- Universität Leipzig
- HTWK Leipzig
- Freie Universität Berlin
- IU Internationale Hochschule
- FernUniversität in Hagen
Learning level from your context
Complete profile details for school type, grade and region or degree programme and semester feed into the prompt. Without enough education context, the fallback remains neutral and follows your input.
Level 2
Lower secondary
Grades 5 to 7. Textbook-style language, technical terms with a short explanation. Curriculum-based by region and school type.
Level 3
Upper secondary / exams
Grades 8 to 13. Higher-secondary level with full technical vocabulary. Exam-syllabus oriented. LaTeX for complex formulas.
Level 4
University
Academic precision. STEM vocabulary at university level. LaTeX for all mathematical expressions. Degree programme and semester in the prompt.
Pedagogical context: the Zone of Proximal Development (Vygotsky, 1978) describes learning with appropriate support; it is not evidence of Quanta product effectiveness.
Why Quanta AI cards differ from ChatGPT output
“Quanta uses education context from your profile when it is complete. If it is missing, the prompt stays neutral and follows your input. With no Bloom selection, no fixed target level is imposed; the path writes for the content and context. You can explicitly select levels 1 to 6, which then become binding targets. In the default source-first path, Quanta fetches real full text, generates from it and verifies grounded cards verbatim against their sources. If no usable source is found, no flashcards are created; a file or a URL is the way.”
What Quanta AI Set does differently, and why
The demo shows a versioned, document-bound Quanta state rather than invented dashboard metrics.
Physics · Radioactivity
Flashcard
Exam prompt
How do mass number and proton number change in alpha decay?
How certain was your answer?
CorrectUnsureWrong
Evidence linked directly to the card
1 / 4 · Switch subject
Free-form chat output
- — The learning goal and level must be described in each prompt
- — Source status depends on the chosen model and prompt
- — The output is not yet a scheduled study set
- — Checking against existing cards requires a separate workflow
- — Moving the output into a study app remains an extra step
Quanta AI Set
- — No fixed Bloom target without a selection; selected levels are binding
- — Level from profile context, otherwise neutral and input-led
- — Distractor validation: plausible MC wrong answers (enforced pedagogically)
- — Duplicate avoidance against a limited set of existing cards
- — Native FSRS-6; scheduling starts after the first rating
- — Integrated AI explanation and adaptive tutor features
- — Source transparency in the source-first path; without a suitable source no flashcards are created
Bloom levels 1 to 6, selectable for the learning goal
With no selection, no fixed Bloom target is imposed; the generation path writes for the content and context. When you select individual levels from 1 to 6, exactly those become binding targets: from remember and understand through apply, analyse, evaluate and create.
Anderson, L. W. & Krathwohl, D. R. (2001). A Taxonomy for Learning, Teaching, and Assessing. Addison Wesley Longman.
Proximal development, why level adaptation is decisive in learning theory
An AI-generated flashcard set at university level is a poor fit for a student in grade 9. When complete education context exists, Quanta passes school type, grade and region or degree programme and semester into the prompt. If those details are missing, the neutral fallback tells the AI to infer level and terminology from the specific input without guessing an education stage.
Vygotsky, L. S. (1978). Mind in Society: The Development of Higher Psychological Processes. Harvard University Press.
Distractor validation for multiple choice, enforced pedagogically
Poorly constructed multiple-choice questions often make wrong answers obvious. Quanta therefore instructs the model to check distractors for plausibility before output. This is a mandatory prompt step, but not a guarantee of error-free options. Haladyna & Downing (1989) provide the methodological context for plausible distractors.
Haladyna, T. M. & Downing, S. M. (1989). A taxonomy of multiple-choice item-writing rules. Applied Measurement in Education, 2(1), 37 to 50.
AI explanations and adaptive tutor features
In a card study session, the user can request a contextual AI explanation. The separate Quanta Tutor uses cards as exam concepts, asks adaptive follow-up questions and provides structured feedback including improvement steps and a model answer. Socratic deepening questions are one question type, not the only response form.
Chi, M. T. H. et al. (2001). Learning from human tutoring. Cognitive Science, 25(4), 471 to 533.
Native FSRS-6 integration, no manual import
Quanta saves generated cards directly in the study system. Once you rate a card for the first time, FSRS-6 calculates stability, difficulty and the next review date. In the open comparison run by the open-spaced-repetition community, FSRS-6 reaches a log-loss of 0.3460 on 349,923,850 reviews from 9,999 collections (retrieved on 5 September 2026); log-loss measures prediction error, lower is better, and that table carries no row for SM-2. The peer-reviewed paper behind it, Ye et al. 2022, reports 220 million behaviour logs and a 12.6% improvement over the state of the art. Both are results of those datasets, not a blanket product effect.
Ye, J., Su, J., Cao, Y. (2022). A Stochastic Shortest Path Algorithm for Optimizing Spaced Repetition. ACM KDD, doi:10.1145/3534678.3539081.
Frequently asked questions
Do I have to upload a document, or is a topic name enough?
For AI Set a topic name and an optional subtopic are enough, with no document required. Quanta finds suitable sources and uses the source-first path by default. If no usable source is found, no flashcards are created; a file or a URL is the way. AI Scan instead works from your uploaded document.
How specific should the topic be?
The more specific, the more precise the cards. "Organic chemistry" produces a broad set. "Nucleophilic substitution SN1 vs SN2" produces focused, in-depth cards. For complex topics we recommend several sets with different subtopics.
What if I have not filled in a profile?
Without enough education context, Quanta uses a neutral fallback: the AI follows the level and terminology of your specific input without guessing a school type or degree programme. For targeted adaptation, add degree programme and semester or grade, region and school type to your profile.
Can I steer the focus of the set?
Yes, through controlled selection instead of free text: the more precisely you phrase the topic, the narrower the set, and Bloom levels 1 to 6 set the cognitive depth as a binding target. There is deliberately no free context field, because freely typed instructions would rank above the evidence in the source-based prompt.
Which AI provider is behind the generator, and where is the processing done?
Quanta uses a European AI provider. AI processing runs through Mistral AI SAS, based in Paris: text functions go via the provider's EU endpoint, while reading files and voice recordings goes via the provider's standard endpoint, for which it gives no specific processing location; the basis is the standard contractual clauses. The provider contractually excludes training on the content we transmit. The provider stores inputs and outputs for 30 rolling days for abuse detection. Card generation is a text function and therefore runs on the regional EU endpoint api.eu.mistral.ai. Accounts, study content and results are stored in region europe-west3, that is Frankfurt am Main. From our side, no account data goes to the AI: no names, no email addresses, no user identifiers, no IP addresses. What a file you upload or a voice recording contains is up to you.
Is the AI flashcard generator included in Starter?
Free forever: 1 topic, no more than 100 cards in total, FSRS, and exactly one AI generation run over the account lifetime. The regular generator offers 40–100 cards from text, a link, or up to 10 files; the guided onboarding entry offers 10–50 from text or files. The onboarding run counts as that one lifetime run; Starter has no monthly AI quota. Further generations require Quanta Essential and share its monthly allowance of 300 AI cards.
AM
Amos Matzke·Founder & Managing Director, Full-Stack Architect · former MINT-EC student·April 2026
Your first AI set from a topic
Enter a topic. Use profile context or stay neutral and input-led. After the first rating, FSRS schedules the review.