Meet the expert

Meet the expert

Puya Sharif

Partner, expert in applied AI and ML

Puya is a Partner at Analysys Mason and co-founder (and chair) of Modulai, the Sweden-based AI and machine learning (ML) consultancy that became part of Analysys Mason in October 2026. Since 2018 he has helped grow Modulai into a team of more than 30 ML engineers. The team has delivered over 150 end-to-end AI projects and put more than 150 models into production for clients ranging from start-ups to Fortune 1000 organisations, including Klarna, Nordea, EQT and Scania.

His work covers the full lifecycle of applied AI, from finding where ML creates business value to designing, building and deploying production systems. His technical focus includes generative AI and agents, computer vision, neuro-linguistic programming (NLP) and predictive modelling. He has particularly deep experience in financial services, especially fraud and credit risk. He likes problems where off-the-shelf models fall short and the solution needs custom architectures, careful evaluation and an honest look at what the data can support.

Before Modulai, Puya built ML systems for high-volume, real-time decisions in Nordic fintech. At Klarna, he designed, engineered and deployed the company’s first ML-based fraud detection system. At Qliro he built the company’s first credit decision engine and its credit risk models, covering everything from the data infrastructure to decision-flow design and statistical modelling.

Puya holds an MSc in Theoretical Physics from Stockholm University. His research there developed quantum game protocols, which use quantum states as a resource to raise expected utility in-game-theoretic settings. That grounding in physics and mathematics still shapes how he works: rigorous about what a model can and cannot do, and pragmatic about what it takes to run it reliably in production.

Outside work, Puya reads extensively on the science of consciousness, from the free energy principle to integrated information theory, where physics, neuroscience and ML meet.