• autobiogrphic memory
• embodied conversational agents
• episodic memory
• graph memory model
• large language models
• long term memory
• nosqldatabases
• semantic memory

In the work we present an innovative approach to enhancing Embodied Conversational Agents (ECAs) by integrating advanced memory management techniques using Large Language Models (LLMs) and NoSQL databases. The core of this research lies in addressing the limitations of current LLMs, particularly their fixed context length, which hampers long-term conversation coherence. The proposal introduces a novel architecture leveraging the Retrieved Augmented Generation (RAG) paradigm, enabling efficient retrieval and integration of historical conversational data.

Key innovations include the development of a hierarchical memory system that mimics human memory stratification into sensory, short-term, and long-term categories. This system ensures contextually relevant and semantically enriched interactions by dynamically updating and recalling episodic and autobiographical memories. By utilizing document and graph storage NoSQL databases, the proposed system can efficiently manage complex and evolving data structures, crucial for personalized user interactions.

This research aims to push the boundaries of current conversational AI capabilities, providing a robust framework for developing more intelligent and human-like ECAs, ultimately contributing to advancements in artificial intelligence and human-computer interaction.