I build ML that has to run outside the lab: in hospitals, on factory floors, on machines that could hurt someone. PhD (Doctor Europaeus), 7+ years across Italy, Germany, Austria, Switzerland. Chartered engineer, Italian register.
Recently tech lead on a medical-imaging foundation model for a European clinical research institute (NDA): self-supervised ViT pretraining on multi-GPU A100, competitive with published domain backbones. Owned the whole stack, from the sharded data pipeline and cross-source standardization of heterogeneous clinical data to a one-command reproducible eval harness, plus the GDPR/EHDS data path and the European cloud/HPC sizing.
In parallel, building an on-premise agentic AI product for regulatory and compliance workflows, where data residency and auditability rule out cloud LLM APIs. Funded by a public R&D innovation grant (ACN/Invitalia). If you need agents somewhere an API key isn't allowed to go, I've lived in that problem.
Before: ROS navigation and perception for mobile robots deployed at industrial scale, 3D perception and grasping at Roboception, MSCA PhD on real-time multimodal ML around people.
Looking for clinical/medtech AI, agentic systems in regulated domains, or robotics perception. The model is the easy par job.
This year I got my PhD (focused in robotics) and walked the camino del norte (a pilgrimage in Spain).
I am trying to transition from academia to industry without success (for now) and several threads here on HN remind me I am not alone in this.
Amazing work on the PhD. I know a guy who worked on a PhD, then moved on to industry in a research position and then started his own firm. Reach me if you want to connect. He's in SF
As a phd student in robotics I drink a lof of coffee. Unfortunately, as always, the title was waaaay bigger than the content.
Missing clear sources and got mad about the regression analysis part *where's my p-value?.
Quite interesting self-hosted tools with overlapping features.
Actual can also periodically sync with your bank to have your most recent transactions on the app. it works with read-only tokens.
Remote: Yes | Willing to relocate: Yes
Availability: Contract/freelance (B2B, EU invoicing) now; open to permanent.
Technologies: Python, C++, PyTorch, self-supervised learning (DINO/ViT), foundation models, multi-GPU distributed training (DDP, A100), medical imaging, computer vision, agentic AI / on-prem LLMs, ROS1/ROS2, Jetson/edge inference, Docker, Kubernetes, AWS/HPC, GDPR/EHDS
Résumé/CV: https://francescovigni.com | https://portfolio.francescovigni.com | https://linkedin.com/in/francesco-vigni
Email: [email protected]
I build ML that has to run outside the lab: in hospitals, on factory floors, on machines that could hurt someone. PhD (Doctor Europaeus), 7+ years across Italy, Germany, Austria, Switzerland. Chartered engineer, Italian register.
Recently tech lead on a medical-imaging foundation model for a European clinical research institute (NDA): self-supervised ViT pretraining on multi-GPU A100, competitive with published domain backbones. Owned the whole stack, from the sharded data pipeline and cross-source standardization of heterogeneous clinical data to a one-command reproducible eval harness, plus the GDPR/EHDS data path and the European cloud/HPC sizing.
In parallel, building an on-premise agentic AI product for regulatory and compliance workflows, where data residency and auditability rule out cloud LLM APIs. Funded by a public R&D innovation grant (ACN/Invitalia). If you need agents somewhere an API key isn't allowed to go, I've lived in that problem.
Before: ROS navigation and perception for mobile robots deployed at industrial scale, 3D perception and grasping at Roboception, MSCA PhD on real-time multimodal ML around people.
Looking for clinical/medtech AI, agentic systems in regulated domains, or robotics perception. The model is the easy par job.