Gianluigi Vitale — Enneper AI

Research focus

Systems research from numerical paths to model behavior.

Gianluigi studies machine-learning systems, LLM serving infrastructure, accelerator portability, TPU kernels and numerical behavior, ML benchmarking, inference reliability, and LLM safety evaluation. He founded Enneper AI to make it easier for people who began independently to meet and build focused research projects together.

Portfolio Google Scholar ORCID GitHub OpenReview LinkedIn

Selected research

Published · MLSys 2026DriftBench Ongoing systems workDeepSeek V4-Flash on TPU Work in progressDSA kernel for TPU Active researchBit-level TPU MXU characterization Ongoing collaborationLong-context genomics Manuscript under reviewFactorial ablation of jailbreak framing B.Sc. thesis in progressDSProfiler

Publications & manuscripts

PublishedDriftBench: Measuring and Predicting Infrastructure Drift in LLM Serving Systems Under reviewDecomposing One Professional-Framing Pipeline: Which Components Shift LLM Safety Boundaries?Under reviewNatural Language Is Flatland for Machine Intelligence: AI Systems Should Decouple Their Reasoning Substrate from Human Language

Background

Education

B.Sc. in Computer Engineering, Universitas Mercatorum — coursework complete; DSProfiler thesis in progress, expected March 2027.

Current role

Systems Engineer at the Archivio di Stato di Pistoia, Italian Ministry of Culture, designing retrieval and public-consultation systems over archival data.

Earlier engineering work

Web Developer at Informamuse S.R.L. from October 2020 to April 2021, working across HTML, CSS, JavaScript, PHP, Laravel, Vue.js, MySQL, and end-to-end delivery.

Academic service

Program Committee member for MLSys 2027; Ethics Reviewer for NeurIPS 2026; Artifact Evaluation Committee member for SOSP 2026 and MLSys 2026.

Research infrastructure

Experience across Google Cloud TPU v4/v5e/v6e, NVIDIA and AMD accelerators, JAX/Pallas, PyTorch, vLLM, SGLang, and TensorRT-LLM.