Pierpasquale Colagrande

Artificial Intelligence Engineer at Blimp.ai

R&D for Blimp.ai's flagship computer vision sensor for urban monitoring; optimized codebase to reduce complexity and increase inference speed; achieved a 130% speed increase on NVIDIA Jetson Nano (3–7 FPS) and a 1286% increase on NVIDIA Jetson Orin Nano with DeepStream (7–97 FPS); developed and optimized AI models for object detection and tracking; improved people counting accuracy by 26% with hyperparameter tuning and by 30% upgrading from MobileNetV2+SORT to Ultralytics YOLOv11-Small; achieved 63% mAP50-95 on detection of additional objects; trained and deployed image classification networks; achieved 92% staff member classification; created, cleaned, and prepared datasets; reduced offline sensors by 65% via a robust input module using GStreamer; enabled a single sensor to handle multiple cameras without FPS loss; created an updater tool and a GUI for remote sensor management; developed REST APIs for client-specific features.

Data Scientist at Lutech

R&D AI projects; AI-driven PoC for MRI analysis to assist in neurodegenerative disease diagnosis; fine-tuned deep learning vision models (CNNs, ViTs) for detection and classification on MRI images and multimodal datasets using Hugging Face; AI-driven hydraulic plant automation and optimization; designed and developed an intelligent pump automation system leveraging Reinforcement Learning and Physics-Informed Deep Learning; created algorithms to convert hydraulic plant schematics into Deep Neural Network architectures via modular building blocks; implemented and trained custom neural networks for predictive, adaptive and efficient plant management; delivered Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) solutions; implemented vision systems with Vertex AI and Google Cloud Platform; built CI/CD pipelines using Docker, Kubernetes, and Mia-Platform.