Large Language Models (LLMs) have significantly advanced knowledge retrieval by enabling more accurate semantic understanding, particularly in Retrieval-Augmented Generation (RAG) systems. Traditionally, knowledge is stored as static text, and combining information from diverse sources remains a human-driven task.

This thesis explores an innovative approach by evaluating the concept of knowledge retrieval through LLM agents within a Multi-Agent Knowledge Graph. A primary focus is on the potential for LLMs to exchange and integrate knowledge autonomously, thereby enhancing retrieval efficiency and accuracy. A secondary focus of this research is the exploration of user- or case-specific perspectives within the graph structure, facilitating semi-automatic configurations of agent teams based on relevant contextual filters. Such context-based filtering enables the dynamic formation of agent setups tailored to specific retrieval tasks, enhancing both relevance and efficiency in knowledge access.

This study will implement a Multi-Agent Knowledge Graph to support knowledge sharing among LLM agents, evaluating the effectiveness of such a framework in optimizing retrieval across diverse knowledge domains. An AI use case library from an industry company will serve as a testbed for understanding how Multi-Agent interactions can facilitate more adaptive and efficient retrieval processes.

Research Questions

• Can knowledge exchange within an LLM Multi-Agent framework improve knowledge retrieval capabilities?
• How can a Multi-Agent-based Knowledge Graph be implemented and applied to an industry use case library?

Prerequisites

• Strong programming skills in Python
• Basic understanding of LLMs and knowledge retrieval systems

Literature

[1] Exploring Large Language Model based Intelligent Agents: Definitions, Methods, and Prospects

[2] Large Language Model based Multi-Agents: A Survey of Progress and Challenges

[3] Agentic Systems : A Business Perspective