Title: Architecture and prototype of a lightweight AI-powered personal learning assistant using open source small language model and multi-modal retrieval augmented generation

Authors: Shilpi Taneja; Siddhartha Sankar Biswas; Bhavya Alankar; Harleen Kaur

Addresses: Department of Computer Science and Engineering, School of Engineering Sciences and Technology, Jamia Hamdard, New Delhi-110062, India ' Department of Computer Science and Engineering, School of Engineering Sciences and Technology, Jamia Hamdard, New Delhi-110062, India ' Department of Computer Science and Engineering, School of Engineering Sciences and Technology, Jamia Hamdard, New Delhi-110062, India ' Department of Computer Science and Engineering, School of Engineering Sciences and Technology, Jamia Hamdard, New Delhi-110062, India

Abstract: In a rapidly changing world where AI is now taking a centre stage in how we interact with various systems, we propose an AI-powered personal learning assistant to help learner gain better understanding of the open educational resources (OER). This assistant transforms the way learners engage with OERs that may contain images, video lectures, text based documents, etc., by interacting with them in natural language on their own local machines without any additional subscription costs. The assistant has multimodal capabilities and uses techniques like multimodal retrieval augmented generation (RAG) and open source small language and multimodal models. Detailed architecture and technology stack for implementation is presented in the paper. A prototype of the proposed assistant is developed and optimised by evaluating different models and adjusting the retrieval chunk-sizes. The assistant is hosted on local machine using open source tool, which makes this architecture cost-effective, light weight and ensures privacy.

Keywords: multimodal large language model; MM-LLM; retrieval augmented generation; RAG; LlamaIndex; open education resources; personalised learning; learning assistant; open source language models; large language models; LLMs; small language models; SLMs; AI in education.

DOI: 10.1504/IJTEL.2026.152750

International Journal of Technology Enhanced Learning, 2026 Vol.18 No.2, pp.144 - 167

Received: 15 Jul 2024
Accepted: 22 Nov 2024
Published online: 09 Apr 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article