MSc in Computer Science and Engineering, Politecnico di Milano (110/110, 2026). Visiting researcher at Northeastern University's Institute for the Intelligent and Networked Systems, working on GPU infrastructure sharing for AI-RAN.

• 🔍 Open to ML infrastructure / systems engineering roles
• 🧩 Open-source contributor: PyTorch ExecuTorch
• 🏆 Innovation Award, NSF SpectrumX Data & Algorithm Competition

MSc Thesis · Northeastern University, joint effort with ANTLab (Polimi) · Sept 2025 – Apr 2026 Supervised by Prof. Antonio Capone (Polimi) and Prof. Michele Polese (Northeastern)

• Extended NVIDIA's DRA Kubernetes plugin with a novel mig-ci device type, enabling dynamic MIG Compute Instances (SM isolation with shared GPU memory)
• Patched vLLM for MIG UUID handling; benchmarked with GuideLLM across three MIG configurations on Northeastern's Open6G OpenShift testbed
• Recovered ~33 GiB of stranded GPU memory: 6.3× KV-cache increase, ~75× higher LLM throughput at peak concurrency, zero measurable 5G NR throughput degradation
• Full thesis text: Politesi
• Related publication: N. Mohamadi, R. Prasad, A. Vacca, et al., ARISE: AI-RAN Infrastructure Sharing and Elastic Orchestration, submitted IEEE INFOCOM 2026 (under review). Contributed the infrastructure layer — the mig-ci DRA extension — achieving ~4× users at peak concurrency, zero 5G degradation, 60% power savings vs. baselines. Code not yet public due to INFOCOM's publication constraints.

AI infrastructure & GPU systems

• mig-ci-metrics — Metrics collection tooling built for the mig-ci DRA device type, supporting the thesis benchmarking pipeline. (Python)
• nv-gpm-metrics — Lightweight GPU/MIG metrics reader (NVML), built as a faster alternative to dcgmi while monitoring GPU behavior for thesis experiments. (C, Python)
• vllm (fork) — Patched for MIG UUID handling as part of the thesis benchmarking pipeline. (Python)
• executorch (fork) — Working fork for PR #19053: extended the Apple MLX delegate to support integer types for aten.bitwise_not, enabling correct lowering of integer bitwise ops on Apple Silicon. (C++, Python)
• libVersioningCompiler (fork) — Upgraded this hot-swap-compiled-functions library from legacy LLVM to LLVM 18/19: rewrote the CMake build system and modernized compiler pass interfaces. (C++, LLVM, CMake)

Systems & low-level

• menshen-open-nic — Port of the Menshen programmable-switch project to the OpenNIC platform. (SystemVerilog)
• CNN-HLS — Hardware description of a handwritten digit recognition CNN via High-Level Synthesis. (C)

Machine learning coursework

• ANNDL-Challenges — Artificial Neural Networks and Deep Learning coursework challenges. (Jupyter Notebook)
• RecSys-2023-polimi — Recommender Systems challenge — collaborative filtering, book recommendation domain. (Jupyter Notebook)