Tommaso Mannelli Mazzoli

• Working on an AI chatbot and document-intelligence applications: OCR, hierarchical retrieval-augmented generation (RAG) and knowledge graphs. • LLM-as-a-judge evaluation and agentic AI applications, including MCP servers. • Local inference of open-weight language models on NVIDIA DGX Spark. • Tools: Microsoft Azure AI Foundry, Microsoft Copilot, Azure DevOps.

• Designed and implemented metaheuristic and hybrid algorithms (LNS, CMSA, tabu search, simulated annealing) for bus driver scheduling with complex break rules. • Developed MIP and MIQP formulations with Gurobi, CPLEX and SCIP. • Built synthetic and realistic benchmark instances from public-transport timetables (GTFS data). • Compared algorithms through systematic computational experiments and Instance Space Analysis; investigated time-series clustering of bus networks. • Publications in JAIR, ICAPS, GECCO

3 months Research stay, supervised by Christian Blum

3 months Research Stay supervised by Kate Smith-Miles

As prescribed by an agreement between University of Florence and UCM, I was a visiting scholar for a period of 6 months, from September 2019 to February 2020.

Thesis in Combinatorial Optimization, "The Quadratic Assignment Problem (QAP): metaheuristic approaches".

Thesis in Differential Algebra: Integration in finite terms, why does "not exist" the primitive of e^(x^2)?