I am a Senior Machine Learning Engineer at Thomson Reuters based in Zurich / Zug, Switzerland.
Working at the intersection of AI and cybersecurity, I focus on both securing AI systems and applying agentic AI to defensive security challenges.
- Autonomous Vulnerability Remediation: Closed-loop agent environments where models triage and fix code vulnerabilities against deterministic verifiers.
- Agent Harness Engineering: Low-level harness architecture, runtime sandboxing, and loop engineering.
- AI Security & Defensive Tooling: Red teaming, prompt injection defenses, and practical guardrails.
- Offensive Security Foundations: Hands-on pentesting, network exploitation, and practical security audits.
codeflow-solver— A graph optimization engine that models vulnerability dataflow graphs as a Set Cover Problem to find the minimal set of code changes needed to remediate security findings.- I am also preparing a few other open-source tools around agent harnesses and security workflows that I will be releasing soon.
I collect structured research notes, paper summaries, and guides covering Machine Learning, AI Security, and Cybersecurity in my Obsidian Vault.
Beyond the screen:
- Swimming
- Specialty coffee
- Custom mechanical keyboards
- Escape rooms
- Cooking
Get in touch:
- Email: gianfranco.romani@outlook.it
- Check/download my resume here