As a Machine Learning Engineer I conducted in-depth research on generative AI for 3D content, guiding project direction in this emerging field.I'm currently developing a data augmentation pipeline to create synthetic data with the goal of improving the reliability of perception models in Autonomous Driving (AD). This pipeline leverages DL models to create 3D objects and scenes to create novel AD scenarios.Moreover, I also contribute to MLOps tasks,having deployed MLFlow (experiment tracking) and Jenkins pipelines (CI/CD), improving the efficiency of our R&D team.

As an AI Data Engineer at Bosch, I have had the opportunity to work on topics related to autonomous driving, mainly contributing to many challenging data-driven tasks that support the development of Artifical Intelligence tasks. More specifically:• Developed Python scripts to validate and transform several terabytes of vehicle sensor data (e.g., RGB, LiDAR), targetedfor DL model development.• Designed and implemented an Apache Airflow pipeline to automate data processing workflows.• Created a Python pipeline to process, store and display DL model metrics on Apache Superset and Voxel Fiftyone.• Improved a C++ sensor data streaming tool stability by addressing memory leaks and concurrency issues.

As an assistant researcher on the PAStor project, I designed and implemented a non-intrusive middleware for Deep Learning services on High-Performance Computing infrastructures. This system deploys storage tiering to accelerate DL models’ training performance and decrease the I/O pressure imposed over the, frequently used, parallel file system. It leverages from existing storage tiers of supercomputers, as well as the I/O patterns of DL solutions to improve data placement across storage tiers.The developed solution accelerated DL training on the Frontera supercomputer by up to 28% (Tensorflow) and 37% (Pytorch), and reduce I/O calls on the Parallel FS by up to 56%.Additional contributions:• Published 2 papers on accelerating DL training, in collaboration with the BigHPC and PAStor projects and the TACC.• Collaborated with other HPC/AI researchers, by integrating a C++ middleware with Pytorch, through pybind11.

I have collaborated towards improving a GCP based IoT solution for fleet management atlarge scale. My work was primarily focused on building data pipelines to achieve scalable and distributed processing of the fleet's generated data. Furthermore, I worked with many Google Cloud Platform technologies.