Legal AI Assistant — KG-Enhanced LLM for Italian Law
Conversational assistant for the Italian legal domain combining RAG with Hybrid Search and a Knowledge Graph-enhanced LLM.
A selection of the projects I built during my years of study, research and work: from multi-agent systems to RAG pipelines, from Knowledge Graphs to LLM fine-tuning.
Conversational assistant for the Italian legal domain combining RAG with Hybrid Search and a Knowledge Graph-enhanced LLM.
Multi-agent system for automatic email management based on CrewAI and an MCP Server as the communication layer.
Evolution of the AgentMail project: native Gmail and Google Calendar integration, vector storage on Qdrant and automatic event creation.
Multi-agent system (9 agents, 3 scenarios) that analyzes PDF resumes and semantically matches them against 100,000 job offers on Qdrant.
Link-prediction task on the decentralized social network BlueSky: classic heuristics and GNN models on a graph of 7,175 nodes.
Study of the MathQA paper and fine-tuning of Mistral 7B with 4-bit quantization (NF4) and LoRA, published on the Hugging Face Hub.
Autogen conversational agents computing a compatibility index between resumes and job offers scraped from LinkedIn.
Study and implementation of advanced RAG techniques (Dense X, RAG Fusion) in collaboration with the company ALTILIA.
Development of a model for the detection and segmentation of Transmission Towers (TTs) and Power Lines (PLs).
Insurance cross-selling prediction: forecasting whether a health-policy customer will also subscribe to car insurance.
Flask web application on Spark and RDDs to query a movie-review dataset in a distributed way.
Python implementation of Adaptive Large Neighborhood Search combined with Simulated Annealing on the Knapsack Problem.