Krishna Reddy Atchi 2 3 ( July - September ) The Large Language Models LLMs provides excellent natural language understanding multistep reasoning and content creation capabilities For example for complex decision making problems conventional single agent systems have limitations that are intrinsic for the system where everything is done within a single structural constraint planning memory management execution and verification This results in hallucination failure to keep context in memory at a higher timescale failure to decompose task and inadequate adaptability in dynamic environment The idea is introduced in this paper of the MultiAgent LLM Framework for Autonomous Decision Systems MALFADS which fragments the cognitive domains into 4 specialized agents Planner Agent Memory Agent Execution Agent Evaluation Agent Agents work independently of each other have a clear assignment and communicate via structured coordination layer Incoming tasks are broken down into atomic subtasks by the Planner Agent persistent contextual knowledge stored in a vector database as part of the Memory Agent data processed by invoking tools and executing them in the Execution Agent and validated outputs and computed confidence scores in the Evaluation Agent The communication protocol in the architecture is dynamic and allows agents to communicate by passing messages which are contextaware MALFADS outperforms all other single LLM RL agent and memoryaugmented baselines on a 50000sample multidomain dataset in terms of decision accuracy 97 task success rate 95 and mean response time 13s Copyright (c) World Journal of Pharmaceutical Seiences. All rights reserved 2026