Project Goals
This project aims to apply Large Language Models (LLMs), a new frontier in Generative Artificial Intelligence (GenAI), to industrial asset management through real-time process monitoring. We investigate and design novel solutions for LLM-based conversational multi-agent system architecture to orchestrate and dynamically distribute tasks among specialized LLM-powered agents. Our approach enables efficient task execution and intelligent decision support. Our goal is to design new software frameworks to allow the development and evaluation of no-click applications that support professionals in real-time monitoring of oil and natural gas production platforms. The project is maintained by the Universidade Estadual de Campinas (UNICAMP).
Multidisciplinary Approach
To advance LLM-powered conversational multi-agent cognitive systems, the project outlines an architecture comprising multiple autonomous agents, each equipped with cognitive modules, that collaborate to extract insights, enhance operational efficiency, and reduce cognitive load on human operators. Each agent consists of five main modules: Planning, Hybrid Memory, Execution, Perception, and Reasoning. Our solution accounts for short- and long-term memory, with distinct memory types used for different purposes within the agent, such as working memory, episodic memory, reflective memory, semantic memory, and procedural memory. Our solution also uses a shared memory space among all agents, serving as a blackboard context for the multi-agent system.
Funding & Partners
This research is financially supported by the Company Petróleo Brasileiro S.A. (Petrobras). The opinions, hypotheses, and conclusions or recommendations expressed in this material are the authors’ responsibility and do not necessarily reflect the views of the Company and other agencies.
