DnyanSetuAI

Research & Technology

What is actually built, filed, and running — described in verifiable terms.

This page exists for one audience in particular: reviewers, technical evaluators, and anyone who wants to check our claims against the code and the filings, not just the pitch. Every figure below is either a government-issued copyright receipt or a description of code that runs today.

PRADNYAVARDHINI — the recall-scheduling algorithm

PRADNYAVARDHINI (प्रज्ञावर्धिनी — "that which cultivates wisdom-intelligence") is the algorithm that decides when a student should see a question again, inside the SAQ (spaced active-recall) engine. It replaces one-size-fits-all schedules with a per-student calculation, run after every "I Recall" / "I Forgot" answer.

The formula

Retention(t) = e^(−t / (τ × stability)) — an Ebbinghaus exponential decay curve. t is days since last review; τ is a fixed 1-day time constant; stability is a per-student, per-question multiplier in the range 0.5–2.0.

How stability is computed

From the student's own last 10 attempts on that question: recall rate → variance → consistency → a stability value between 0.5 (volatile) and 2.0 (consistent). No population averages — every student gets their own curve.

Binary signal only

The only input is is_recalled ∈ {0, 1} — "I Recall" or "I Forgot." No multiple choice, no partial credit, enforced at the database layer with a CHECK constraint, not just in the UI.

What it outputs

An integer interval, clamped to 1–90 days, converted into a next_review_at timestamp and written back to the student's memory-state record for that question.

In plain terms: two consecutive right answers on a question a student answers consistently will push that question further out; a student who is inconsistent on a question sees it again sooner. The multiplier is derived, not assumed.

Scope of the claim, stated precisely

PRADNYAVARDHINI's implemented formula has two components: a fixed time constant and a computed per-student stability factor. It does not currently incorporate a knowledge-graph adjacency term or a separate affective/emotional-state input into that specific calculation — those are different, ongoing parts of the wider platform (see below), not additional multipliers inside this formula today. We'd rather describe the algorithm narrowly and correctly than describe it broadly and imprecisely.

Where the wider science stands today

DnyanSetuAI's architecture is grounded in six peer-reviewed cognitive-science foundations. An internal implementation audit — conducted before any of the filings below — scored the platform's code against 26 specific claims drawn from those foundations. The honest result: 19 fully implemented, 4 partial, 3 not yet built, roughly 73% aligned at the time of that review.

Active Recall (Roediger & Karpicke, 2006)

Fully implemented. Binary-only interaction, no recognition-based testing, architecturally enforced.

Cognitive Load Minimization (Sweller)

Fully implemented. One question at a time, no scroll, no clutter, offline-capable.

Affective Filter (Krashen)

Fully implemented as Mann-Setu, the separate emotional-support layer described on our Trust Center — a Hindi-first guided journey run before study begins, not a variable inside the PRADNYAVARDHINI formula.

Ebbinghaus Forgetting Curve

Algorithm implemented and, since this review, wired into the live queue processor: every recall interaction now computes and stores a personalized next_review_at date.

Spaced Repetition (Leitner)

Partial. Weak-question detection and per-student scheduling are live; proactive push-notification reminders are not yet built.

Mastery Learning (Bloom's 2 Sigma)

Partial. Per-chapter accuracy is tracked and shown to students; hard prerequisite gating between chapters is not yet enforced.

Filed intellectual property

Six works have been filed with the Copyright Office of India (copyright.gov.in), Form XIV, Class IX, all under DnyanSetu Learning Ecosystem LLP. Diary numbers are checkable directly at the Copyright Office's public status portal.

DnyanSetuAI Active Recall SAQ Engine (Source Code)

Diary SW-19548/2026-CO — the binary-recall system architecture: frontend capture, async queue processing, blind-spot detection.

PRADNYAVARDHINI Personalized Spaced Repetition Algorithm (Source Code)

Diary SW-19561/2026-CO — the Ebbinghaus-based scheduling algorithm described above.

Universal Probing Tool (UPT)

Diary SW-19619/2026-CO — the diagnostic questioning tool underlying our assessment flows.

DnyanSetu Universal Master Compiler & Examination Anchor Engine

Diary LD-19590/2026-CO — a five-phase educational content processing pipeline that turns raw syllabus material into structured recall anchors.

The Recall-Relate-Represent Doctrine

Diary LD-19596/2026-CO — the pedagogical framework behind how a student is asked to work with each concept.

Structure for Fulfilment (SFF) & Dattak Yojana Mentorship Protocol

Diary LD-19602/2026-CO — the mentorship and commitment structure behind our field-based counselling work.

All six were filed the same day, 25 April 2026, and share sequential diary and receipt numbers — a single filing session, not six disconnected claims.

Why this page is written this way

We would rather a reviewer find a narrow, accurate claim here than a broad one they have to go looking for a caveat on. If you are checking our work — diary numbers against the Copyright Office, or the formula above against the code that runs it — we want that check to come back clean. Questions about any of the above: dnyansetuai@gmail.com.