Konstantinos Christos Rigatos

At NIKI Digital, I designed and delivered a production agentic RAG system with hybrid retrieval, cross-encoder reranking, trust decay, and evidence-based abstention. What began as internal research became a commercial product adopted by two additional clients.

I also proposed a confidence-weighted knowledge-graph extraction architecture for a large pharmaceutical database, using a few-shot LLM to supervise a lower-cost non-LLM student model. I strengthened production delivery through architecture decision records, typing, CI, testing, Docker, Postgres, nginx, and Linux infrastructure.

My independent work includes PyTorch physics-informed neural networks for constrained optimisation in conformal field theory and uncertainty-quantified regression using Bayesian ridge models. I value results that hold up to validation, including clearly reporting limitations when a model cannot resolve them.

Before moving into AI/ML engineering, I spent a decade in theoretical physics research across the University of Santiago de Compostela, KITS, and the University of Colorado Boulder. I hold a PhD from the University of Southampton and have published 16 peer-reviewed papers, including five single-authored publications.