A student reads an explanation twice, recognises every word, and still cannot solve the next question. A teacher rewrites the same idea for three different levels of understanding. These are the moments that make us ask a harder question: what should a useful AI explanation be built around?
Gyanis is moving towards Next-Concept Modeling (NCM) as a direction for future research and product development. Our interest is in how emerging approaches to text generation could support clearer, more responsive learning experiences. This is a statement of direction, rather than an announcement that NCM is already powering Gyanis.
What is Next-Concept Modeling?
Next-Concept Modeling is a research framework for generating text through progressively more detailed representations. The October 2026 work by Hermann Kumbong, Ishan S. Khare and colleagues describes a hierarchy of discrete codes. A model generates that hierarchy level by level, predicting the codes within each level in parallel, before a decoder reconstructs the text. Read the original paper.
Think of a designer establishing the broad shape of a building before refining its rooms and details. That analogy helps explain the progression; it does not mean the model thinks like a person. In this research, “concept” refers to a learned representation, not necessarily a named school topic or a human-readable lesson outline.
Why the research has our attention
The researchers report promising generation-speed results against specific diffusion-model baselines. Those are results from their experimental setup, not measured improvements in Gyanis. They do not establish better classroom outcomes or prove that every existing language model should be replaced. The research team's explanation provides useful context.
For us, the interesting question is practical: could a different way of generating explanations make it easier to deliver useful support at the moment a learner needs it? That possibility deserves careful investigation.
Personalisation starts with the learner's difficulty
Imagine a learner who writes 1/3 + 1/4 = 2/7. A helpful response should identify the misunderstanding about equal-sized parts, choose a suitable example, and give the learner another chance to try. A longer answer alone will not do that.
Our ambition is an experience that connects the learner's attempt, the explanation and the next opportunity to practise. NCM could become one component in exploring that ambition. It does not automatically supply a learner model, curriculum knowledge, assessment reliability or the ability to choose the next topic. Each of those needs its own design and evaluation.
This distinction matters for parents and teachers: Next-Concept Modeling is about a generation approach. Deciding what a student should learn next is a separate educational problem.
Where this direction fits within Gyanis
Gyanis Flow brings together learning activities such as practice, mindmaps, planning and handwritten-answer work through Express. These experiences create practical questions for future research. Can an explanation remain consistent when expanded? Can feedback stay grounded in the student's actual working? Can a learner use it to solve a new problem independently?
In Express, for example, a handwritten answer may contain a useful method alongside a mistaken step. Any future generation approach would still need reliable interpretation of that work and checks on the feedback. NCM itself is not handwriting recognition.
Gyanis Vibe and Gyanis Aura broaden the ecosystem through tutoring and wellbeing support. Their human relationships and responsibilities remain central. Moving towards NCM does not mean that a text-generation architecture replaces a teacher's judgment or a coach's care.
What meaningful progress would need to show
Before describing an NCM-based experience as an improvement, we would want evidence against today's approach. A sensible evaluation would examine factual accuracy, consistency, response time, cost and how well the explanation fits the task. It should also test whether students can apply what they have learned without simply copying the answer.
For learners in India, the UAE and Malaysia, curriculum and language differences also matter. Success on one benchmark cannot stand in for testing the subjects, wording and contexts people actually use.
We have not announced a release date or published Gyanis-specific NCM benchmark results. Our direction is clear: move towards Next-Concept Modeling with the learner's experience as the test of whether the technology belongs in the product.
Common questions
Is Gyanis already using NCM?
This article announces a future direction. It does not claim a deployed NCM implementation in Flow, Vibe or Aura.
Did Gyanis invent Next-Concept Modeling?
No. The research discussed here belongs to the authors credited in the linked paper. Gyanis is expressing its own direction in relation to that work.
Will NCM automatically improve personalisation?
No. Personalisation depends on appropriate learner context, instructional choices and evaluation as well as the generation method. Any benefit must be demonstrated.
The goal is an explanation that helps someone take their next step with understanding. That is the standard we want our future work to meet. Explore Gyanis and follow our updates as this direction develops.
Research context: October 2026. The examples above illustrate our aspirations and evaluation questions; they are not reports of completed NCM experiments.