◦ Developed an intelligent conversational agent designed for company briefings and corporate
communications using LangChain framework for building robust language model applications.
◦ Implemented comprehensive security measures focusing on prompt injection prevention and data leakage
protection to ensure safe deployment in enterprise environments.
◦ Conducted extensive testing across different stages of the AI pipeline using LangGraph for workflow
orchestration and LangSmith for monitoring and evaluation of model performance

Explored optimization strategies for Retrieval-Augmented Generation (RAG) by comparing Late Chunking and Early Chunking approaches. Evaluated the impact of static and dynamic segmenting on retrieval and generation performance, highlighting trade-offs in efficiency, context preservation, and scalability. Achieved insights into designing adaptive chunking strategies for enhanced RAG workflows.
Skills: Retrieval-Augmented Generation (RAG) · LLM · RAG · Pattern Recognition · Pytorch · LLMs

This project leverages Convolutional Deep Neural Networks (Conv-DNN) and Vision-Language Models (VLM) to predict pedestrian behaviors in real-time. The model integrates spatial visual data and semantic context to anticipate future movements of pedestrians, enhancing the safety and responsiveness of autonomous driving systems. By analyzing environmental cues and pedestrian dynamics, this solution contributes to safer navigation in complex urban environments.
Skills: Autonomous driving Systems · Deep Neural Networks · Computer Vision · Vision Language Models · Pattern Recognition · Pytorch

This project focuses on developing computer vision systems capable of detecting anomalies in images or video data. It involves training models to identify patterns that deviate from normal visual behavior, which is crucial in applications such as industrial inspection, medical imaging, and surveillance. The project explores unsupervised, semi-supervised, and deep learning-based approaches to accurately detect rare and subtle anomalies, even with limited labeled data.
Skills: Convolutional Neural Networks (CNN) · Deep Learning · Computer Vision · VLMs · Pattern Recognition

This project explores how Large Language Models (LLMs) can be enhanced to perform complex reasoning and planning tasks. It focuses on evaluating and improving the model’s ability to break down multi-step problems, make logical inferences, and generate coherent action sequences toward specific goals. The project integrates techniques such as chain-of-thought prompting, tool use, memory management, and modular architectures to simulate cognitive-like decision-making in LLMs.
Project done at Cineca using the supercomputer Leonardo,