DAVIDE MODOLO

AI Engineer, R&D

OpenCity Labs (Startup) · Remote · August 2025 - Present

Architected a multi-tenant conversational AI system on the open-source Cheshire Cat AI framework, developing and releasing multiple open-source plugins. Started exploring Model Context Protocol (MCP) to standardize tool execution, building a Client-Server proof of concept for autonomous appointment booking. Built a core NLP pipeline: fine-tuned multilingual Transformer-based NER models for PII anonymization, a spaCy-based sentiment analysis module optimized for fast CPU inference, and an output validation layer for automated translation.

AI Engineer Intern

Eurecat Technology Center · Barcelona, Spain · April 2024 - June 2024

Implemented a token-level uncertainty estimation framework for planning tasks by analyzing log-probabilities, benchmarking across GPT-4o and locally quantized models (Llama 3.1 via llama.cpp). Developed a RAG system with gte-large embeddings to retrieve context-aware user preferences, significantly reducing model hallucination in recurrent decision-making. Prototyped a multimodal agent integrating LLaVA for visual reasoning tasks.

Master's Degree in Artificial Intelligence Systems

University of Trento | Trento, Italy | September 2021 - March 2025

Thesis: "Exploring the Use of LLMs for Agent Planning: Strengths and Weaknesses" - Research on Large Language Models applied to automated planning and decision-making scenarios.

Bachelor's Degree in Computer Science

University of Trento | Trento, Italy | September 2017 - June 2021

Thesis: "Healthy Plus - Redesign and evolution of an Android application for monitoring healthy lifestyles" - Complete redesign and modular development of a health tracking mobile application.

Designed a multi-stage deep learning model to classify Lung Ultrasound images based on a 0-to-3 illness score. Developed 3 components: multi-class frame classifier, uncertainty detection model, and similarity module using 47k frames from 14 real patients.

Implemented 4 Deep Learning models for simultaneous Intent Detection and Slot Filling tasks. Achieved 13% improvement on SNIPS dataset in slot filling task. Fine-tuned BERT and ERNIE, built Bidirectional LSTM and Encoder-Decoder models from scratch.

Built and evaluated a deep learning model for Unsupervised Domain Adaptation using ResNet34 with custom adaptation layer. Achieved +11.53% improvement over baseline non-adapted model on Adaptiope dataset using 3rd and 4th-order statistics.

Designed autonomous agent solution for delivery problems using Belief-Desire-Intent framework. Developed and tested both single-agent and collaborative multi-agent environments across 7 challenges with 2 different approaches.

Provided a solution to Closest Pair of Points problem in N dimensions with parallel implementation. Implemented both Bruteforce and Divide et Impera solutions using Message Passing Interface (MPI) in C. Tested with up to 250M points using 1 to 80 CPU cores in 4 different configurations on the University's cluster.