Luca Lillo

WORK HISTORY

EDUCATION

ABOUT LUCA LILLO

I design and build production-grade AI systems where reliability, auditability and human oversight are not optional features, but core architectural principles.My work focuses on integrating large language models into structured, governed pipelines rather than treating them as standalone decision-makers. I specialise in retrieval-augmented generation (RAG), multi-step orchestration and validation layers that ensure outputs are traceable, explainable and aligned with real-world constraints — built on a stack that includes PyTorch, FastAPI, LangChain, LlamaIndex, Docker, Kubernetes, Terraform and AWS.A central element of my approach is Human-in-the-Loop (HITL): AI systems should assist, not replace, human judgment. I design workflows in which models generate insights, scenarios and structured analyses, while human experts retain control over interpretation, validation and final decisions.I am particularly interested in high-risk and regulated environments, where AI must operate within clear boundaries of accountability, compliance and governance. My work aligns with the principles of the EU AI Act, adopting a risk-based approach and embedding human oversight, transparency and control mechanisms directly into system design.I also conduct independent research on LLM agent governance. My paper \"Can LLM Agents Care About the World?\" introduces a formal control framework for world-directed agent welfare and is available as a preprint on ResearchGate.