Exploring Multimodal AI Frameworks for Real‑Time Decision Making in Edge Devices

Abstract

The rapid advancement of Artificial Intelligence (AI) and edge computing has driven the demand for intelligent systems capable of real-time decision making under limited computational resources. In particular, multimodal AI, which integrates heterogeneous data sources such as visual, audio, and sensor signals, plays a crucial role in enhancing contextual awareness and decision accuracy at the edge. This study aims to explore and conceptualize a multimodal AI framework that supports real-time decision making on edge devices while addressing challenges related to resource constraints, data privacy, and decision transparency. The research adopts a qualitative literature review approach, employing a Systematic Literature Review (SLR) method to analyze relevant studies published between 2018 and 2025. Data were collected from reputable academic databases and analyzed using thematic content analysis to identify key architectural components, fusion strategies, optimization techniques, and privacy-preserving mechanisms. The findings indicate that hybrid multimodal fusion, combined with model compression, dynamic inference, and federated learning, significantly improves efficiency, privacy protection, and explainability in edge-based AI systems. This study contributes a comprehensive conceptual framework that can guide future development and deployment of adaptive, efficient, and trustworthy multimodal AI solutions for real-time edge intelligence applications.