Master Thesis: “Development of a Modular Multi-Agent System Architecture for Enhanced Flexibility and Scalability”
Author: Greshma Shaji
Supervisors: Ziqi Wang, Prof. Enrique Zuazua, Prof. Frauke Liers
Date: November, 2025
Large Language Models (LLMs) are increasingly used in autonomous systems but face challenges in long-term reasoning, coordination, and scalability. This thesis introduces a hybrid, modular Multi-Agent System (MAS) architecture designed to overcome these challenges by combining LLM-driven planning with deterministic orchestration and real-time observability.
The proposed system integrates a PlannerAgent responsible for goal decomposition and workflow generation, alongside persistent WorkerAgents executing specialized tasks. A Redis-based commu- nication layer enables dynamic coordination and plug-and-play scalability. The architecture supports self-evaluation, adaptive replanning, and transparent monitoring through integrated feedback loops and a Streamlit-based dashboard. This combination achieves a balance between autonomy, determinism, and explainability, paving the way for more reliable agentic systems.
See all details at:
