Marginalia
Viktor Lofgren
My name is Viktor. I’m a Swedish software engineer and hypertext enjoyer. Marginalia is a website I’ve built. It’s really almost a bunch of websites on a common theme. If you find yourself clicking a link and ending up on a page that looks completely different, that’s just how things are.
July 2026
Top Words
- marginalia
- search
- index
- software
- years
- grant
- crawler
- astrolabe
- internet
- query
All posts (137)
One of the big ambitions for the search engine this year has been to enable searching in more languages than English, and a pilot project for this has just been completed, allowing experimental support for German, French and Swedish. These changes are now live for testing, but with an extremely small corpus of...
I’ve been using the CoPilot plugin for IntelliJ on and off for the last few years, and while initially pretty enthusiastic, I’ve come to first disable it and then delete it altogether along with JetBrains’ local AI-completions, and generally felt this has been an improvement in productivity and a reduction of...
How you engage with the world changes how you experience the world, and how the world experiences you. A snarky and cynical approach, by its default assumption that things are shit, or if they are not yet shit will inevitably turn to; such an approach will give your world a malodorous brownish tint. Granted, snark...
The Marginalia Search index has been partially rewritten to perform much better, using new data structures designed to make better use of modern hardware. This post will cover the new design, and will also touch upon some of the unexpected and unintuitive performance characteristics of NVMe SSDs when it comes to...
As some of the work planned for Marginalia Search this year has been progressing a bit faster than anticipated, there was time to implement an unplanned change. This post details the implementation of a system for detecting when servers are online, to avoid serving dead links and improve data quality, and for...
The most recent change to the search engine is a system that profiles websites based on their rendered DOM. The goal is identifying advertisements, trackers, nuisance popovers, and similar elements. The search engine already tries to do this, but isn’t very good at it because it’s only looking at static code. It...
You wake at 05:30 in the morning, feeling somewhat groggy. Instead of the alarm clock ringing like it normally does, a cheerful hologram appears: “Hi! I’m Kyle, your new alarm clock assistant!” You get dressed as Kyle explains all of the fantastic things he is capable of. You head over to the coffee machine. “Hey...
The search engine has recently gained the ability to index the PDF file format. The change will deploy over a few months. Extracting text information from PDFs is a significantly bigger challenge than it might seem. The crux of the problem is that the file format isn’t a text format at all, but a graphical format....
Some time ago, I migrated the crawler off the okhttp library, to use Java’s builtin HTTP client. This seemed like a good idea at the time, but has led to a fair number of headaches. Java’s HttpClient has one damning flaw, and that that it doesn’t support socket timeouts. Its only supported timeout values are time to...
A problem the search engine’s crawler has struggled with for some time is that it takes a fairly long time to finish up, usually spending several days wrapping up the final few domains. This has been actualized recently, since the migration to slop crawl data has dropped memory requirements of the crawler by...
Page 2 of 14