Projects & research

Development of a Multi-Agent System for Efficient Retrieval-Augmented Generation in Generative AI Systems

My bachelor thesis, awarded a perfect 10/10, proposes a novel multi-agent architecture for Retrieval-Augmented Generation. Instead of a single model, five specialised agents collaborate through a shared state space to resolve complex, multi-step queries that overwhelm traditional single-agent RAG. Built on LangGraph, it adopts the open Model Context Protocol (MCP) as a unified, reliable layer for accessing external tools and data, plus a structured mechanism for error detection, classification, and recovery. Evaluated on MCP-Bench with an LLM-as-a-Judge and rule-based metrics, it delivers higher accuracy, traceability, and robustness than single-agent baselines.