<![CDATA[Retrieval-Augmented Generation (RAG) has received attention in various studies, but research specifically addressing adaptive chunking strategies for Indonesian-language Informatics learning documents remains very limited. This study aims to design a hybrid adaptive chunking system that routes each document section to an appropriate chunking strategy based on structural signals detected at the preprocessing stage. This study used a Design and Development Research (DDR) approach through the stages of document analysis, system architecture design, and expert validation involving three experts in Informatics and Natural Language Processing (NLP). Data were collected through structured expert review instruments and scenario walkthrough sessions. The results showed that rule-based structural detection was able to reliably distinguish heading, narrative, list, and code block sections, supported by a confidence-based fallback mechanism. The conclusion of this study affirms that hybrid adaptive chunking plays an important role in maintaining the semantic coherence of learning materials in RAG systems. These findings contribute to the development of adaptive information retrieval studies and broaden understanding of RAG design aligned with pedagogical needs in the Indonesian-language academic context. The implications of this study include the provision of a reusable design framework for Indonesian-language technical documents and practical guidance for developers of educational AI systems.]]>
