Founder & CEO · LineSpotting · Sweden · Born 1974
He goes in where delivery has stalled, puts a number on the value within two weeks, then builds the system that makes the improvement stick. For three decades that craft was consulting and turnaround. Now it is AI — on the network, at the edge, and in the P&L.
What buyers of distressed or underperforming businesses purchase is improvement potential. That is the same product David has sold since the nineties: enter a stuck operation, find the bottlenecks, quantify the value before anyone writes a plan. At the Swedish Social Insurance Agency (Försäkringskassan) that two-week discipline identified at least 150 million euro per year in recurring savings, plus a portfolio of mobile and portal modernisation proposals.
The degree on the wall is a Master in Film & Television (Bond University) — not computer science. The substance is elsewhere: assembler in the mid-eighties, RAS remote access for Deloitte’s offices two decades before “remote work” was a slogan, and today edge ML and AI-run infrastructure in production. Six Sigma Black Belt since 2009. CISSP. IPMA Level B. The method is formal; the delivery is hands-on.
Since 2022 he has run LineSpotting, an AI incubator with 25+ parallel projects across energy, IoT, water, waste and critical infrastructure. The firm scaled to 85 people. After he applied AI to the company’s own operations, the same portfolio runs with a single person at the centre. Not a client pilot — his own accounts.
Reuse is the design rule: a concept that works in one company should roll into the next. That is why AI IT Department ships as a finished package — site, PDF and PowerPoint from one source — Meraki-API-driven operations, roughly half the headcount to run the same estate, measured security posture up about 17%. It is the same logic behind Alutbildningar.org (making adoption repeatable for leaders and teams) and behind ai.ze.mt, a prompt optimiser and cost router that sends each job to the cheapest model that can still do the work, with token, cache and budget tracking per job.
OCCDEC is not a slide. It is acoustic event detection in production: YAMNet TFLite on Jetson Nano at the edge, 16 kHz in overlapping three-second windows, ~180 ms inference per chunk, 521 sound events mapped to fifteen threat classes. Each sensor calibrates its own acoustic fingerprint in thirty seconds. Uncertain events go to a human review queue; approved annotations retrain the model and roll out over-the-air. False alarms down about 85%; alarm out in under a minute. Full MLOps loop — not a diagram.
Discovery workshops that put leadership and engineers in one room. Use cases prioritised by effect, feasibility and effort — Six Sigma, not opinion. Implementation with vendor selection, technical validation and programme control. Tooling that makes the next deployment cheaper than the last. Travel four days a week is normal; delivery experience across Sweden, Australia, Estonia, Finland, Norway, Germany and France.
The four-slide talk you came from is one product in that portfolio: how to run IT infrastructure with AI so headcount and security move in the right directions at once. Sources, method notes and a worked NIST CSF 2.0 / ISO 27001 example sit next to it. The figures −50% and +17% are engagement averages from his own client work — not a vendor benchmark.