Paul Tonelli ML Infrastructure Engineer
Cell: [USE E-MAIL]
Address: Paris
Email: paul@tonel.li
Currently senior consultant at Akvize where I mainly manage the infrastructure and AI aspects of audits.
- Skills
-
Hardware and Software Architecture
Kubernetes, Docker/Containerd, CRI-O, multi-cloud high availability (bare-metal, AWS, GCP, Azure), storage (Minio, S3, HDFS), cost optimization, Terraform, Talos, Ansible
GPU/ML Infrastructure
CUDA, cuDNN, nvidia-docker, Alibaba GPUShare, GPU scheduling and observability, large-scale data preprocessing pipelines, ML analysis pipeline co-design
Management
Built and managed an infrastructure team for 6+ years (recruited 3 senior engineers). SCRUM methodology, roadmap definition, technical recruitment.
Programming
Python3, C++, zsh/bash
Monitoring and Security
Grafana, Prometheus (custom GPU exporters), ELK, Loki. Security risk identification, environment isolation, incident management, GPG, SSH, SSL/TLS.
- Experience
-
Akvize (ex CloudixIO)Senior Infrastructure Consultant - Paris - 2024-present
- 40+ technical audits across critical sectors (trading, insurance, critical services)
- System architecture expertise, AI strategy and cybersecurity
- Deep Learning and GenAI audits, model protection strategies
- Technical recruitment: conducting interviews for senior engineering positions
- Talos Linux, Terraform, multi-region Kubernetes deployments
HeuritechHead of Infrastructure/Manager, co-founder - Paris - 2014-2024
- Built and managed the infrastructure team for 6+ years (recruited 3 senior engineers)
- Technical bridge between infrastructure and ML research teams
- Co-designed ML analysis pipelines and optimized data loading with prefetching
- Scalable ML infrastructure processing 5M+ images daily
- GPU infrastructure: nvidia-docker, CUDA/cuDNN, GPU scheduling (NVIDIA GPU Operator, Alibaba gpushare-extender)
- Built containerized architecture (LXC → Docker-Swarm → Kubernetes)
- 99.9% availability using multi-cloud resilience strategy (bare-metal, AWS, GCP)
- Optimized infrastructure costs while scaling GPU compute capacity
- Automated deployment: reduced server rack deployment time from 3 days to 2 hours
LogilabDevOps - Paris - 2012-2014
ISIRResearch Engineer - ISIR (UPMC), Paris - 2011-2012
- Work on the Sferes_v2 C++ framework
ISIRPhD Student - ISIR (UPMC), Paris - 2008-2011
- Courses given: Introduction to Robotics (UPMC), Newtonian Physics (Nanterre University)
Single Buoy MooringsInternship - Monaco - 2006
- Analysis on weather data
Résid. Arts & MétiersCommunity work - Aix-en-Provence - 2004-2006
- System admin for a +300 computer network (User support, firewall and hardware upgrades, cable management system, linux administration)
- Education
-
PhD at ISIR (Sorbonne University)
Thesis Title: Relationships between Synaptic Plasticity and Coding Regularity in Neuro-Evolution (PDF)
MSc in Artificial Intelligence and Decision (Sorbonne University)
Report Title: Verticalizing a bioloid robot
École Nationale des Arts et Métiers
Major: Information and Knowledge Management
Project: Speech Synthesis Project for Mute People (link in french) - Fifth Garangeat prize in 2007. LINK - Publications
-
On the Relationships between Synaptic Plasticity and Generative Systems
Tonelli, P. and Mouret, J.-B. (2011). Proceedings of the 13th Annual Conference on Genetic and Evolutionary Computation. Pages 1531--1538. - Best paper of the Generative and Developmental Systems (GDS) track. PDFUsing a Map-Based Encoding to Evolve Plastic Neural Networks
Tonelli, P. and Mouret, J.-B. (2011). Proceedings of IEEE Symposium Series on Computational Intelligence. Pages 9-16. PDFArtificial Evolution of Plastic Neural Networks: a few Key Concepts
Mouret, J.-B. and Tonelli, P. (2011). DevLeaNN 2011: A Workshop on Development and Learning in Artificial Neural Networks. Pages 32-37. PDF - Interests
- Research Reward-modulated STDP, neural simulation models, continual learning
- Sports Running, Sailing (skipper in Mediterranean Sea and windsurf)
- Home Automation Home Assistant, motion detection, Z-wave, lircd, low power servers (thin-itx, pico-psu, Raspberry Pi)...
- Programming my Github