Giorgio Roffo PhD

Head of Artificial Intelligence & Manager

Equixly — Agentic API Security · Verona, Italy

Equixly provides continuous, AI-agent-driven API penetration testing: agentic AI hackers testing at machine speed within modern CI/CD workflows, trusted by leading European banks, insurers, and payment/mobile providers. I lead the AI team and direct the company's applied AI research, taking self-hosted LLMs and agentic systems from research to production for automated API security testing — under strict ML protocols, rigorous data curation, and state-of-the-art evaluation.

• AI-powered security agents: a production conversational assistant (trained LLM agent with RAG, security layers, and custom-script capabilities); an agent that repairs OpenAPI/Swagger specification files like a code agent with tool calling and RAG; an exploitation agent with workflow escalation, tool calling, and memory management; LLM-based penetration-test report generation with RAG for business-impact analysis; and an EU AI Act–compliant customer knowledge base.
• LLM training & fine-tuning: PEFT/LoRA fine-tuning of self-hosted models; SFT and RL alignment; research on new self-attention mechanisms (infinite and linear attention).
• Efficient serving (vLLM): KV cache, prefix caching, prefill/decode stage balancing, batching for parallel loads, and speculative decoding (EAGLE-3, d-Flash, MTP) with speculated-token tuning.
• Quantization: FP8 and NVFP4 quantization of open-weight models with calibration; Quantization-Aware Training (QAT) for FP8 models and LoRA adapters.
• Agentic AI & RL: RL training with tool calling; modern agentic workflows (Reflexion, Self-Refine, Voyager, CodeAct, MetaGPT, ChatDev, Mixture-of-Agents); penetration-testing agents for analysis and exploitation.
• Security ML pipeline: schema-driven generation, stateful dependency-aware fuzzing, vulnerability-oriented testing, LLM-guided learning, and multi-agent systems, with benchmarks built from stateful, relationship-aware Docker service endpoints.
• AI governance & management: per-feature compliance documentation and risk categorization under the EU AI Act; team planning, issue assignment, and deadline management in collaboration with the Product Manager and CTO.

Technical Lead & Senior AI Research Scientist

Cosmo Intelligent Medical Devices (IMD) · Milan, Italy

Led AI R&D for GI Genius, the first FDA-cleared AI system in gastroenterology, deployed worldwide.

• Endoscopy vision models: built Polyp Sizing, Bowel Preparation Score Estimation, and Polyp Characterization with ResNet and Vision Transformers (ViT) via full fine-tuning — FDA-cleared and deployed worldwide on GI Genius.
• Voice-based clinical reporting: technical lead of a clinical-reporting application in two flavors — a multimodal LLM + TTS pipeline and a standard LLM pipeline — with ASR via Whisper large-v3 and variants.
• Data curation: DINOv2 embeddings for anomaly detection and clustering, surfacing mislabeled samples to clean training data.
• AI governance: led the team's technical due diligence for third-party partners (assessing and onboarding external partner applications).
• Full lifecycle: research, prototyping, and deployment as regulatory-compliant microservices; published in Medical Image Analysis (2025) and at MICCAI 2024.

Principal Investigator / Research Lead

Mind Vision Labs (UCL spin-off) & Toyota Research Institute · London, UK

AI research on autonomous-driving technologies with Prof. Nilli Lavie (head, UCL Attention and Cognitive Control Lab), partnering with Toyota Motor Europe's AI Division.

• Driver attention / distraction detection with Vision Transformers (ViT), applying clustering and anomaly detection on embedded video frames to flag driver distraction.
• Research on linear self-attention and Transformers for large-scale sensor data, improving autonomous-driving perception pipelines.
• Research output: Self-Attention and Beyond the Infinite — a linear-attention Transformer with infinite self-attention, submitted to BMVC 2026.

Senior Research Scientist (Post-Doc)

University of Glasgow · Glasgow, UK

Three-year post-doc funded by an EPSRC investment of £355,564; managed a complex multi-university research project.

• Deep learning and attention models for social signal processing with Prof. Alessandro Vinciarelli; taught one-third of the undergraduate Artificial Intelligence course (2019–2020).
• Managed and mentored MSc/PhD students to graduation (multimodal depression analysis, autism-spectrum video analysis, attachment disorders), with peer-reviewed publications in IEEE, ACM, and BMVC venues.
• Established a Glasgow–Stanford collaboration, supported by an appointment as Visiting Postdoctoral Scholar at Stanford University (Vision & Learning Lab, Prof. Silvio Savarese, 2019).
• Weekly research collaboration with Heriot-Watt University on the EPSRC-funded SoCoRo project (2017–2019) — socially competent robots for autism research — recognized with a SICSA award for excellence in research collaboration.

Research Associate / PhD Candidate

University of Verona · Verona, Italy

• PhD in Computer Science with Doctor Europaeus certificate (May 2017). Thesis: Ranking to Learn and Learning to Rank: On the Role of Ranking in Pattern Recognition Applications (Prof. Marco Cristani); 400+ citations from the thesis work.
• Created Infinite Feature Selection (Inf-FS), connecting graph theory and feature selection by formulating the selection of features as a path on a graph — published at ICCV 2015, ICCV 2017, and in IEEE T-PAMI 2020.
• Authored the Feature Selection Library (FSLib) for MATLAB — 11,000+ downloads — MathWorks Outstanding Contribution Award (2016) and MathWorks Research Summit invitations (Newton, USA).
• Awarded the Cooperint International Programme (CIP) grant, University of Verona (2015).

Research Associate

Italian Institute of Technology (IIT) · Genova, Italy

• Computer vision and user authentication with Prof. Vittorio Murino (head, PAVIS Lab) and Dr. Loris Bazzani: part-based models for pedestrian detection and tracking, pose estimation, and characterizing human behavior from social-media data.
• Three-month research internship at the University of Glasgow, School of Psychology (Prof. Frank Pollick): ML classification of expert vs. novice CCTV-operator eye movements — first-author paper at CIARP 2013.