ESCARGOT: an AI agent leveraging large language models, dynamic graph of thoughts, and biomedical knowledge graphs for enhanced reasoning

Abstract

In this paper, we introduced ESCARGOT, a novel AI agent framework designed to mitigate hallucinations in LLMs by integrating dynamic reasoning through the Graph of Thoughts (GoT) framework with robust knowledge retrieval mechanisms. ESCARGOT dynamically generates a GoT based on specific queries, informed by the LLM’s strategy formulation, and executes this plan through a Python-based AI agent. By leveraging Cypher-based extraction from knowledge graphs and vector database queries, ESCARGOT ensures the retrieval of relevant, contextually accurate information while minimizing hallucinations. ESC