Technical articles on AI systems, agentic development, and EU R&D.
Human approval is not a weakness in agentic systems. It is the control layer that makes AI useful for high-stakes workflows.
Lessons from building HEDGE ExpertAI: vector search is powerful, but domain discovery improves when semantic retrieval, keyword scoring, and ontology signals work together.
Production-ready AI is less about impressive prompts and more about data contracts, observability, reliability, human workflows, and deployment discipline.
The difference between using AI as a tool and orchestrating AI agents as collaborators — how agentic development workflows changed how I ship production software.
What I learned deploying energy management systems in real manufacturing facilities and FabLabs — from sensor hardware to production Docker stacks.