ProRAG: Towards Reliable and Proficient AIGC-Based Digital Avatar

The concept of a Virtual Human represents an advanced interactive interface that bridges users with digital information, offering an increasingly realistic experience. Recent breakthroughs in Large Language Models (LLMs) and AI-Generated Content (AIGC) have significantly improved the lifelike nature of virtual humans, making them increasingly indistinguishable from real humans. However, this rapid progress raises significant concerns regarding the ethical implications and the reliability of virtual human interactions, particularly in high-stakes, domain-specific scenarios where factual accuracy and trustworthiness are paramount. In response to these challenges, we introduce ProRAG, a novel framework designed to enhance the trustworthiness and reliability of digital avatars. ProRAG combines domain-specific LLMs with innovative strategies to address key challenges such as hallucinations, computational inefficiency, and context stability. Our approach integrates a multimodal knowledge base, consisting of textual, visual, and auditory data, to improve retrieval accuracy and content consistency. Furthermore, ProRAG supports multimodal digital human interactions, facilitating voice, visual, and text communication, which ensures high trust for critical applications. By leveraging adaptive data representation techniques, ProRAG resolves the “Lost in the Middle" challenge, enhancing hallucination suppression and promoting structured knowledge integration. This framework is designed to be scalable and versatile, demonstrating its potential across diverse domains such as education, cultural preservation, and legal consultation, while ensuring the generation of reliable, context-aware content in mission-critical decision-making environments.