Co-PIs: Le Duy Dung ([email protected]); Nguyen Phi Le ([email protected]); Nguyen Quoc Viet Hung ([email protected])

The project aims to transform how personalization technologies understand and respond to human preferences. Current recommender systems, whether based on collaborative filtering or deep learning, struggle with limited adaptability, opaque reasoning processes, and dependence on static user data. These limitations hinder user trust and fail to meet the growing societal demand for transparency, fairness, and human-centered AI.

This project proposes a new generation of agentic recommender systems powered by large language model (LLM)-based agents with capabilities for reasoning, planning, memory, and self-improvement. By enabling systems to act autonomously, interpret user intent through natural language, and explain their decisions, the research will bridge the gap between algorithmic efficiency and cognitive interpretability. Building on recent advances in LLM and Agentic AI, this work introduces a unified, modular architecture for explainable, adaptive, and ethically aligned personalization.

The core innovation question guiding this project is: How can agentic AI leverage reasoning and explainability to create adaptive, trustworthy, and socially responsible recommender systems? The research will design agentic frameworks capable of continuous learning through interaction and evaluation across dynamic user environments.

The societal challenge addressed is the urgent need for trustworthy AI personalization in domains such as education, healthcare, and sustainable e-commerce. The project’s impact pathway includes developing open benchmarks, explainability modules, and real-world prototypes in collaboration with industry and academic partners. 

Expected outcomes include: 

• Societal impact: Enhanced efficiency, user trust and transparency in AI-driven decisions. 

• Economic impact: Efficient, adaptive personalization that improves digital service value. 

• Environmental impact: Reduced data and computational waste through context-aware recommendations. 

Ultimately, this research will establish a foundation for human-centered, autonomous, and explainable agentic AI systems that advance ethical personalization in the digital age.