I build AI for regulated decisions, and I write about where the AI hype is right and where it is wrong.
Scroll down
About me
Since 2021
Started as an IT consultancy in Oslo. Today it is DVERSI's founding solution and implementation partner.
Since 2026
The intelligence layer for your business systems, starting with property risk assessment for insurers.
Book: AI Orchestration: The Definitive Guide. Preprint: Darshana: A Six-School Framework for Large Language Model Orchestration and Training.
I founded Hundred Solutions in 2021 and DVERSI in 2026. DVERSI is the intelligence layer for your business systems, built for regulated organisations in the Nordics and starting with property risk assessment for insurers. People make the decisions.
I have spent about twenty years in enterprise technology, more than ten of them leading large IT services delivery across Scandinavia and Europe. Much of that time was inside the ERPs, CRMs and accounting platforms that AI is now expected to work with. I know why these systems break because I used to keep them running.
"Intelligence built for the systems your business already runs on."
Work
Intelligence layer
DVERSI is the intelligence layer for your business systems, built for regulated organisations in the Nordics. It works with the systems you already run, starting with property risk assessment for insurers: each case is checked against the insurer's own rules. People make the decisions.
Your existing systems stay in place
Property risk assessment first
People make the decisions
Solution partner
Hundred Solutions is DVERSI's founding solution and implementation partner. We find the real business problem, design how the work should run, then configure, integrate and support DVERSI for a specific team, with delivery from Norway and India.
Discovery and process design before any tooling
Integration with your existing systems
Rollout, training and support
Publishes
Most organisations bolt AI onto legacy tools without asking why the process exists. This book is a diagnosis before it is an implementation guide. The frameworks come from real production work, and so do the warnings.
Solve first, automate later. Zero-token workflows for the parts of a process that never needed a model. A multi-agent architecture that holds up in production. A way to judge whether an AI business is defensible or only demoable. It is what building real systems for real clients taught me.
Read on Amazon
Paper
In this preprint I map six schools of Indian epistemology onto layers of an LLM pipeline and test each against an equally detailed generic prompt. The main finding is about placement: the same technique won 0% of comparisons at runtime and 82% as a query rewriter. Across 4,400+ generations, all five components beat their controls. It is not yet peer reviewed.
Read the paper83%
Vritti: labelling how it knows
60% when GPT-4o judged
82%
Mimamsa: rewriting the question
67% when GPT-4o judged
93%
Nyaya: deciding when to search
70% fewer searches
71%
Vaisheshika: organising knowledge
vs a generic entity ontology
82%
Vedanta: merging sources
63% when GPT-4o judged
Reads
Speaking
I am most useful in rooms where a decision is being made: a board weighing an AI investment, a leadership team choosing a direction, or a technical group designing something they will live with for years.
Working sessions with boards and leadership teams on where AI fits in their business, and where it does not.
The book behind most of these conversations. Reading it first makes our time together more useful.
Available on AmazonSolve the real problem first, by hand if needed. Automate it only once the solution is proven.
A research session on what classical theories of knowledge can teach us about structuring LLM systems, based on my Darshana preprint.
