• Agentic AI
• Generative AI
• LLM
• Multi-Agent
• Semantic Kernel

This thesis focuses on building an experimental multi-agent system powered by large language models (LLMs), using Microsoft’s Semantic Kernel framework. It explores two orchestration approaches: a collaborative setup, where LLM agents communicate and reason together, and a non-collaborative approach, where agents operate independently to perform assigned tasks. A simple demo application was developed with a FastAPI backend and a Streamlit frontend, enabling communication via HTTP and open communication via WebSocket APIs. The solution includes retrieval-augmented generation (RAG) and external service integration to extend the agents’ abilities. This project was developed during an internship at a company interested in evaluating the Semantic Kernel framework to understand its practical real-world usage and potential to be offered to clients.