AI Can Calculate a Risk. It Can’t Carry One.
AI helped us replace a 150,000 kroner a year licence in a few evenings. What becomes scarce when building gets cheap, and one question every leader should ask at each point of no …
technical product manager at izy · oslo, norway
Saurav Pandey
I build AI that runs in production,
and write about who carries the risk when it fails.
At Izy I lead the backend and the AI layer behind hundreds of thousands of transactions a month: computer vision self-checkout and demand forecasting, live across Norwegian commercial buildings.
Eight years across backend architecture, machine learning and product. I write about AI and accountability, and created the sleep test, a framework first published in Dagens Perspektiv.
Eight years across backend architecture, machine learning and product. I write about AI and accountability, and created the sleep test, a framework first published in Dagens Perspektiv.
Published in
Blog
Coding Is Solved. The Hard Part Starts Now.
Coding is, more or less, solved. That is what people closest to the work are saying now, and I think they are broadly right. Some context before I go further. I spent the last few …
Using AI for recruitment in Norway? Here is how algorithms inherit our biases
AI is changing how companies hire people. It’s happening everywhere, including Norway. Despite Norway’s reputation for equality, NAVIGATE, a research series at UiO shows clear …
EU's AI Act 2024: Safeguarding Consumers or Restricting Tech Growth?
In May 2024, the European Union formally adopted new regulations under the Artificial Intelligence Act that sets out rules for how AI can be sold, used and monitored in the EU and …
What backs this up
Led the team that built AI self-checkout at Izy
Computer vision that recognises a lunch tray and charges for it without a queue, running daily across Norwegian commercial buildings, with demand forecasting for the kitchens alongside it. Most people advising on AI have never had to operate what they recommended at 07:00 on a Monday.
Replaced a 150,000 kroner a year licence in a few evenings
One person, AI in every step from idea to design, build, run and repair, and a human approving every reply before it goes out. Stable in production. A few years ago it would have taken a team and a quarter.
Created the sleep test, published in Dagens Perspektiv
At every point of no return, ask one question: who loses sleep if this goes wrong? If the answer is a team or a process and not a name, you are not ready. A working rule for keeping accountability when AI does the work.
Formal training in AI ethics, two peer-reviewed papers
160 hours on fairness, causal intervention, governance and deployment through Turing College. Published research on anomaly-based intrusion detection and blockchain supply chains (IJAE). Applied to systems that are actually running.
About
I have worked with machine learning since 2017, when my final-year project and first published paper used it to detect network intrusions, years before AI went mainstream. Today I am Technical Product Manager at Izy in Oslo. I lead the production backend for a Norwegian workplace platform and the computer vision and demand forecasting that run on top of it.
Eight years across backend architecture, machine learning and technical product. Nearly all of it on systems that run every day rather than systems that demo well.
What I care about is the part after the model works: whether the data holds up, whether the team can operate it, and whether a named person stands behind every decision that cannot be undone. That last question became the sleep test, and it is why I keep writing.
Taking AI from pilot to production?
I talk with leaders and teams who want AI in real operations without losing a name behind every decision that matters. Talks, workshops and a second opinion on where you are stuck. More on how I work.
Email is fastest, and I answer it myself.
Email is fastest, and I answer it myself.