Research-Backed Adaptive Learning
Every algorithm, every question sequence, and every AI interaction in Alfanumrik is grounded in peer-reviewed learning science and responsible AI principles.
How We Model Learning
Our adaptive engine draws on decades of cognitive science and educational psychology research.
Bayesian Knowledge Tracing
We model each student's mastery as a probability distribution, updated in real-time as they answer questions. This lets us estimate not just what they know, but how confident we are in that estimate.
Bloom's Taxonomy Progression
Questions are tagged across six cognitive levels — Remember, Understand, Apply, Analyze, Evaluate, Create. Students progress upward only when lower levels are mastered.
Zone of Proximal Development
Adaptive difficulty keeps students in the sweet spot — challenging enough to grow, but not so hard they disengage. Based on Vygotsky's foundational learning theory.
Spaced Repetition (SM-2)
The SuperMemo SM-2 algorithm schedules review sessions at increasing intervals, optimizing the transfer from short-term to long-term memory.
Responsible AI for Education
AI in education demands the highest standards of safety, accuracy, and age-appropriateness.
Content Moderation
All AI-generated responses pass through multi-layer content filters ensuring age-appropriate, safe, and educational content only.
Age-Appropriate Responses
Foxy adjusts language complexity, examples, and tone based on the student's grade level and age.
Curriculum Alignment (NCERT/CBSE)
AI responses are grounded in NCERT textbooks and CBSE curriculum. Foxy never teaches content outside the approved syllabus.
Hallucination Prevention
Retrieval-augmented generation (RAG) with curriculum-specific embeddings ensures factual accuracy. Confidence scoring flags uncertain responses for review.
Measuring Understanding, Not Just Answers
Our assessment framework goes beyond right and wrong to understand how students think.
Item Response Theory (IRT)
Each question is calibrated using IRT parameters — difficulty, discrimination, and guessing. This enables precise measurement of student ability independent of the specific questions asked.
Cognitive Load Theory
Quiz sessions are designed to manage intrinsic, extraneous, and germane cognitive load. Question sequencing and hint systems prevent overload.
Error Classification
Wrong answers are categorized as careless errors, conceptual misunderstandings, or misinterpretations. Each type triggers different remediation strategies.
Ethical Data Practices
Student data is a sacred trust. Our data practices reflect that responsibility.
DPDPA Compliance
Full compliance with the Digital Personal Data Protection Act, 2023. Lawful basis for processing, data principal rights, and grievance redressal mechanisms.
Data Minimization
We collect only what's necessary for the learning experience. No unnecessary personal data collection, no behavioral tracking for advertising.
No Advertising
Alfanumrik will never show ads to students. Student data is never used for advertising purposes or sold to third parties.
Parental Consent for Minors
A parent sees a linked child's learning record only after recording their own consent, and can withdraw it at any time.