By day I work in data migration and integration. In my spare time I build things with public data, and use them as an excuse to experiment with AI.
01 Selected work
A thousand years of recorded conflict, a live earthquake catalogue, and a wind forecast checked against real stations. Each one is built around a decision about what its data can honestly say.
An interactive map of Iceland's historical battles and violent conflicts.
The decisionEvery event carries a confidence level, and saga-era places that are uncertain are marked as approximate instead of presented as fact.
A real-time map of earthquakes, volcanic systems and official warnings around Iceland.
The decisionEach region is compared only with its own past, so activity is ranked by how unusual it is there, not by raw count.
Live wind over the Reykjavík capital region, drawn as particles and scored against real observations.
The decisionIt grades itself: every station is checked against the model at its exact coordinates, with bias, MAE and RMSE on show.
02 More work
Every name in the Icelandic name register, broken into the parts it is built from.
AI-drafted meanings are never merged into the cited data. They stay behind an óstaðfest (unconfirmed) badge.
A map-first explorer of actual property sales in the Reykjavík capital region.
The simpler model won: a ridge regression lost to comparable sales on a time-separated holdout, so it stayed out of the product.
The website for an Icelandic nonprofit working to strengthen rural communities.
A theatrical spinning wheel and random picker.
03 About
I spend my working life moving data between systems that were never meant to talk to each other: reconciling shapes, chasing down the records that do not fit, and making sure nothing quietly disappears in transit.
Outside that, I build small things properly. Most of what is here started as a question I could not answer by searching, so I built the thing that answers it. I like projects where the hard part is the data model rather than the framework, and I would rather ship something narrow and finished than broad and vague.
Vindur switches its station markers to model values the moment you scrub away from now, and says so. Iceland Live states that its dispersal runs are simulations before it lists a single one.
In Nafnaval, cited dictionary entries always win. AI-drafted meanings fill the gaps behind an óstaðfest badge and are never merged into the cited data.
Fasteign tested a ridge regression against plain comparable sales on a time-separated holdout. The regression lost, so it stayed out of the product.
04 Contact