Sep 2026 — now
MSc Student, High Performance Computing Engineering
Politecnico di Milano · Milan
Focusing on high-performance computing, numerical methods, parallel and distributed systems, and machine learning.
MSc student · Milan, Italy
Yahia Ghallale
I build and study intelligent systems — from computer vision on edge devices to the infrastructure that makes machine learning faster and more reliable.
Research interests
High-performance computing
Computer vision
Machine learning
Edge inference
Reliable evaluation
Experience
Feb 2026 — Apr 2026
Computer Vision Engineer
Prospecto Srl · Brescia
Developed and optimized real-time computer vision models for smart agriculture and NVIDIA edge devices.
Mar 2024 — Mar 2025
Embedded Software Engineer
GIVI UniBS Motorsport · Brescia
Built ESP32 and STM32 sensor systems for telemetry and real-time data collection during track tests.
Education
2026 — 2028
MSc, High Performance Computing Engineering
Politecnico di Milano · Milan, Italy
2023 — 2026
BSc, Computer Science and Engineering
Università degli Studi di Brescia · Brescia, Italy
2025
Erasmus+ Blended Intensive Programme
From Smart Sensors to Industrial Applications · Covilhã, Portugal
Selected work
GPU Memory Broker
A C++20 broker for GPU memory contention, with priority scheduling, model eviction, and cross-process sharing across CUDA, Metal, and a mock backend.
Repository01
MORPH: Demonstration-Grounded Manipulation World Model
Turns overhead human demonstrations into Panda manipulation rollouts in MuJoCo, then learns an action-conditioned model of the simulated system. Held-out-video evaluation improves arm-motion prediction; object dynamics and outcome ranking remain open challenges.
Repository02
Cross-dataset X-ray classification
Benchmarked five CNNs on patient-level chest X-ray data across three datasets. Reached 95.0% balanced accuracy on RSNA and measured generalization with confidence intervals and paired tests.
Repository03
PyTorch contribution
Submitted a core fix for consistent eager and torch.compile behavior in torch.normal, with tensor standard-deviation validation and an Inductor CPU regression test.
04
Vision at the edge
Deployed an on-device vision pipeline on NVIDIA Jetson Nano with ROS 2 and ONNX Runtime: 95.9% macro-F1 at 19.4 ms per frame.
Prospecto · Computer Vision Engineer05