Joakim
Hedenstedt
Contact
Links
Areas of expertise
- AI Governance and Oversight
- Energy-Efficient AI Inference
- Edge Computing
- Commercialization of Research
- Business Strategy
- Go-to-market Strategy
- Business Development
- Embedded Systems
- Distributed and Decentralized Systems
Biography
Joakim Hedenstedt is Head of Development (Commercialization) in the AICRIT Research-to-Business project at the University of Turku Faculty of Law. He works on taking AICRIT's research to market: an independent second opinion on automated decisions in the public sector, and energy-efficient AI inference.
He has more than 30 years of industry experience in mobile communications and embedded systems, in executive, sales, business development, marketing and engineering roles at Texas Instruments, Mentor Graphics and mobile software companies in the United States, France, Sweden and Finland. He co-founded a startup in financial and digital inclusion, advises startups, and has worked in Rwanda, Nigeria, Zambia, Namibia, Kenya and Tanzania.
He holds an MSc in Engineering from KTH Royal Institute of Technology, Stockholm, and has studied marketing at UC Berkeley.
Research
My research is about AI that organizations can run and audit themselves, locally and without depending on cloud services they do not control, and about how to show that automated decisions follow the rules that apply.
As AI concentrates in a few hyperscale providers, organizations come to depend on them twice: for compute, and for the rules the AI follows. My argument is that capability and governance have to be decentralized together. The work has two interdependent parts:
- Resource-efficient local inference: running capable models within the compute and energy budget of edge hardware, and measuring what that costs.
- Oversight built into the architecture: critic agents, machine-readable rules and provenance running locally, so that a system detects when it cannot meet a rule, escalates to a human and leaves an audit trail. The aim is a system that is compliance-aware, not one that claims to be compliant.
The work also asks what this means for access to AI in markets where centralized resources are expensive, limited or unreliable.