India's first LLM for K-12, built on NEP 2020 and NCF 2023.

Pragya (प्रज्ञा, Sanskrit for "wisdom") is fine-tuned for one job, AI tutoring, on 20,000+ curated teaching samples built from the CBSE curriculum and the teaching approach of NEP 2020 and NCF 2023. It guides children to work out answers for themselves, and maps every answer to the learning goal for their class. Classes 1 to 8 are live; Classes 9 to 12 are planned. Pragya is the model. The YoLearn.ai AI tutor runs on it, and so can your product.

• The policy it is built on: NEP 2020
• The pedagogy it follows: NCF 2023
• Fine-tuned on: CBSE curriculum data
• Teaches in: 12+ Indian languages

How NEP 2020 and NCF 2023 shape every answer.

General-purpose models learn mostly from Western internet data, and they tend to give students the final answer. With Pragya, NEP 2020 sets what it teaches towards, and NCF 2023 sets how it teaches.

• Understanding over rote learning: Socratic, step-by-step guidance that helps a student reach the answer.
• India-centric learning: examples from Indian life and culture, and from the Indian Knowledge Systems that NEP 2020 asks schools to use.
• 21st-century competencies: the 6Cs (critical thinking, creativity, collaboration, communication, character and citizenship) are built into how Pragya asks and explains.
• Cross-curricular connections: links a science concept to mathematics, language, social science or the child's surroundings.
• Stage-wise curricular goals: mapped to the competencies NCF 2023 sets for each stage, so Pragya teaches at the child's level.

Teaching principles: Socratic questioning, inquiry and discovery, reasoning before answers, learning from mistakes, and step-by-step explanations.

Aligned to the NEP's four stages

Pragya adjusts its tone, depth, language and teaching method to each stage of the 5+3+3+4 structure.

• Foundational: Pre-school to Grade 2. Play and real objects, very short replies.
• Preparatory: Grade 3 to 5. Discovery and concrete examples.
• Middle: Grade 6 to 8. Socratic inquiry, from concrete to abstract.
• Secondary: Grade 9 to 12 (coming soon)

Meet Pragya: Yolearn's Large Language Model

Who it teaches

Students in Classes 1 to 8, and their parents, following NCERT: Foundational (Classes 1 to 2), Preparatory (Classes 3 to 5) and Middle (Classes 6 to 8). Subjects: English, Hindi, Mathematics, EVS / Science and Social Science. Ask in English, Hindi, Hinglish and 12+ Indian regional languages, and Pragya replies in the same language.

How it teaches, the NEP 2020 way

• The child thinks first: notice, guess or try before the explanation.
• Understanding over memorising: the why, not just the what.
• Starts from the child's world: home, festivals, games, nature.
• Each step is only a little harder than the last.
• Every question is welcome, and mistakes become lessons.
• A safe space, with no ridicule and no comparison.
• Every answer aims at an NCF learning goal, not at finishing the page.

With every answer

• A clear explanation at the child's level
• The NCF-SE learning goal it serves
• The NEP-NCF qualities it shows
• The NCERT passage it is based on
• The pedagogical style Pragya used

Judged by teachers, not just benchmarks.

YoLearn.ai evaluated Pragya on 5,000 NCERT-aligned questions, judged by a panel of 20 practising teachers. Answers were scored 1 to 5 on four criteria: NEP-style teaching for the stage, correctness, the right level for the class, and the student's language. Teachers preferred Pragya's answer in 92% of cases, and Pragya scored 85% higher than a leading general-purpose model. Every release is evaluated against NEP 2020 learning outcomes.

A language model, fine-tuned to teach.

Pragya is built on a large language model of around 27 billion parameters. Fine-tuning teaches it the NEP-NCF teaching style while keeping what the base model already knows. It was trained on 20K+ curated, quality-checked samples; 70%+ carry an Indian context. Every answer is mapped to the 127 curricular goals and 368 competencies of NCF-SE 2023, and shows its top 3 of 20 NEP-NCF tags.

• Data: concept explanations, worked examples and story problems; Socratic dialogues and misconception correction; activity-based lessons and formative checks; about 20% of samples are in Hinglish.
• Fine-tuning: LoRA fine-tuning of the language layers, with general-knowledge, maths, Hindi and code samples mixed in and a held-out general set tracked throughout, over 10 training stages with a checkpoint after each.
• Selection: each checkpoint was tested on chapters never seen in training, checked for correct learning goals and NEP-NCF tags, and compared head-to-head with the base model. The best checkpoint was chosen for Pragya.

Trained, hosted and governed in India.

A sovereign model for India's classrooms. YoLearn.ai runs the full training pipeline itself and builds the curriculum datasets from CBSE content. Indian teachers review Pragya's teaching behaviour, and every release is evaluated against NEP 2020 learning outcomes. The model and all student data stay on infrastructure in India, so schools, publishers and states that build on Pragya are not exposed to a foreign provider's pricing, policy changes or data terms. Pragya supports the Government of India's goal of self-reliance in AI under the IndiaAI Mission and Atmanirbhar Bharat.

Safe by design

• Stays on the lesson: Pragya teaches only the selected class and subject.
• Built not to invent facts: answers are tied to curricular goals, and every one can be reported for review.
• Points children to trusted adults: for personal concerns, Pragya guides a child to a parent or teacher.

The Pragya roadmap

• Phase 1: LLM for K-8 (Pragya-1, in public testing)
• Phase 2: LLM for K-12
• Phase 3: Subject-specific models for AI tutoring

Access for publishers, school chains and state governments

Tell us what you're building and the grades you serve: Contact@yolearn.ai. Already a Yolearn.ai partner? Pragya already powers your platform.