CognitoLearn: Adaptive AI for Personalized Paths and Mastery

Abstract

CognitoLearn is an AI-driven adaptive learning platform designed to revolutionize personalized education by combining intelligent learning path generation, automated summarization, and conversational mentorship with a robust mastery validation system. Traditional educational platforms largely deliver static, linear content that fails to accommodate individual cognitive states and pacing. CognitoLearn addresses this by leveraging advanced natural language processing (NLP) and generative AI models. The system dynamically generates structured curriculum graphs using fine-tuned T5 models and provides context-aware tutoring via Retrieval-Augmented Generation (RAG). Furthermore, the platform integrates Bayesian Knowledge Tracing (BKT) to probabilistically validate true cognitive mastery rather than relying on heuristic grading, ensuring that learners must genuinely understand prerequisites before progressing. This paper details the architecture, probabilistic methodologies, scaling strategies, and zero-trust validation pipelines that enable this secure, real-time adaptive learning ecosystem, providing an exhaustive evaluation of system latency, BKT accuracy, and RAG contextual integrity.

Introduction

CognitoLearn is an AI-powered adaptive learning platform designed to overcome the limitations of traditional Learning Management Systems (LMS) and MOOCs, which typically deliver static, one-size-fits-all content. Existing platforms often fail to adapt to individual learning styles, knowledge levels, and learning speeds, resulting in reduced engagement and ineffective mastery assessment. They also rely on simple percentage-based grading that measures short-term recall rather than genuine understanding.

To address these challenges, CognitoLearn introduces a Concept-to-Checkpoint Mastery System that combines personalized curriculum generation, intelligent tutoring, and probabilistic mastery assessment. The platform uses a hybrid AI architecture to create customized learning pathways, provide real-time contextual mentorship, and mathematically verify a learner’s understanding before allowing progression to advanced topics.

The system is built on a microservices architecture consisting of a React.js frontend, a Node.js API gateway, a Python-based intelligence engine, and MongoDB databases. The Node.js server manages authentication, routing, and database operations, while computationally intensive AI tasks are handled by Python services using FastAPI. This separation ensures high performance, scalability, and responsiveness.

A key feature of CognitoLearn is Automated Curriculum Generation, where fine-tuned T5 transformer models create personalized knowledge graphs in the form of Directed Acyclic Graphs (DAGs). These graphs organize topics and prerequisites into a logical learning sequence tailored to the learner’s goals and skill level. The roadmap is visualized interactively through React Flow.

For mastery assessment, the platform employs Bayesian Knowledge Tracing (BKT) instead of traditional grading methods. BKT models learning as a Hidden Markov Model and continuously updates the probability that a learner has mastered a concept based on quiz performance. A mastery threshold of 90% is required before unlocking subsequent topics, reducing the likelihood of advancement through guessing or memorization.

To provide reliable AI tutoring, CognitoLearn integrates Retrieval-Augmented Generation (RAG). Educational content is converted into vector embeddings and stored in a semantic search database. When students ask questions, the system retrieves relevant learning materials and combines them with the learner’s current progress before generating responses. This approach minimizes AI hallucinations and ensures explanations remain grounded in verified educational resources.

The platform also includes a Zero-Trust Validation Pipeline that validates and sanitizes all AI-generated outputs before presenting them to users. This mechanism prevents formatting errors, invalid JSON structures, and unreliable responses, improving system stability and trustworthiness.

Performance evaluation demonstrates strong scalability and responsiveness. Standard API requests achieve response times below 100 milliseconds, vector searches complete within 200 milliseconds, and AI-generated learning roadmaps are produced in approximately 3–6 seconds. The RAG-based tutor provides initial responses in under 1.8 seconds, ensuring a smooth user experience.

Conclusion

CognitoLearn bridges the critical and persistent gap between static, one-size-fits-all Learning Management Systems and dynamic, highly responsive intelligent tutoring environments. By architecting a robust MERN stack supported by a dedicated, asynchronous Python AI microservice, the platform seamlessly translates natural language learning goals into structured, hierarchical knowledge graphs with minimal latency. The integration of the Bayesian Knowledge Tracing (BKT) mathematical framework fundamentally elevates the educational assessment process. It ensures that academic progression is statistically tied to true, demonstrated comprehension rather than mere rote memorization or lucky guessing. Coupled with a highly context-aware Retrieval-Augmented Generation (RAG) Chatbot and a Transformer-based content summarizer, CognitoLearn demonstrates that Generative AI can be safely, securely, and effectively harnessed to democratize highly personalized, mastery-based education. The successful implementation of the zero-trust validation pipeline further establishes a blueprint for handling erratic LLM outputs in strict application environments.

References

[1] S. Sarrab, et al., \"Architectural Patterns for Real-Time Adaptive Learning Systems,\" Journal of Educational Technology Systems, 2024. [2] Y. Han, et al., \"Mitigating Hallucinations in Educational LLMs via Retrieval-Augmented Generation,\" IEEE Transactions on Learning Technologies, 2024. [3] X. Liu, et al., \"Automated Construction of Educational Knowledge Graphs (EduKGs) using Transformer Models,\" Artificial Intelligence in Education, 2025. [4] A. T. Corbett and J. R. Anderson, \"Knowledge Tracing: Modeling the Acquisition of Procedural Knowledge,\" User Modeling and User-Adapted Interaction, vol. 5, no. 4, pp. 253-278, 1995. [5] S. Russell and P. Norvig, Artificial Intelligence: A Modern Approach, 4th ed., Pearson, 2020. [6] P. I. Pavlik Jr, H. Cen, and K. R. Koedinger, \"Performance Factors Analysis--A New Alternative to Knowledge Tracing,\" in Proceedings of the 2009 conference on Artificial Intelligence in Education: Building Learning Systems that Care: From Knowledge Representation to Affective Modelling, 2009, pp. 531-538. [7] J. D. Baker, \"The Purpose, Process, and Methods of Writing a Literature Review,\" AORN Journal, vol. 103, no. 3, pp. 265-269, 2016. [8] C. Raffel et al., \"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer,\" Journal of Machine Learning Research, vol. 21, pp. 1-67, 2020. [9] A. Vaswani et al., \"Attention is All you Need,\" in Advances in Neural Information Processing Systems, 2017, pp. 5998-6008. [10] Lewis, P., et al., \"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,\" Advances in Neural Information Processing Systems, vol. 33, pp. 9459-9474, 2020. [11] B. Bloom, \"The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring,\" Educational Researcher, vol. 13, no. 6, pp. 4-16, 1984.

Copyright

Copyright © 2026 Saketharama Badam, Durga Prasad Gundeti, Jayanth Karnati, Rishi Kumar Komuravelli, Juhi Juwairiyyah. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.