andreasthinks.me
[1]
Andreas Varotsis
Data scientist and public safety researcher
I’m Andreas, a data scientist and public safety researcher working to improve the use of data and evidence in government. I enjoy coordinating exciting communities, and working with fantastic institutions to build projects that improve the world, including teaching on public sector innovation at Newspeak House. You can read more about me on my about page here.
If you’d like to discuss professional work or collaborating on a project, please say hello!
Fig. 1drag to rotate
[2]
📮 Blog Feed
I use this blog to journal on what I’ve been learning about and building recently. If you’d like to see more of my formal work, you can see a selection of my papers and writing here.
Thank God for the FOI Act: Strategic Insight Unit releases in the open
Some of my proudest work was done with the Metropolitan Police SIU. I'd sort of assumed much of it would never see the light of day, but it turns out plenty has already been released! So thank god (or at least New Labour) for the Freedom of Information Act, and here are some reports, backed-up for posterity.
When fewer commits don't mean less work
AI coding tools appear to change commit behaviour, which makes commit counts a shaky productivity metric.
I have been vibe coding a lot. Like, a lot a lot…. and it’s been leaving me feeling a bit conflicted. I’ve been committing hard, and feeling tremendously productive. But I…
UK Police Salary Explorer
An interactive look at police pay across ranks from 1999 to 2025, in nominal and real terms.
I made ten AI models play Blood Bowl against each other (in the Discworld)
What happens when you let LLMs loose on a fantasy sports pitch
This is an experiment on collaborating with AI: I wrote most of this blog post by collaborating with my Claw/Hermes Agent, running on a Raspberry Pi. I gave it a topic, we…
Octosphere, an experiment in decentralised science with ATProto
Bridging Octopus and the Atmosphere with lots of vibes
So, I really don’t like the scientific publishing ecosystem. I used to find the idea deeply inspiring and faintly aspirational, and then I had to actually interact with it…
Reflecting on the National Police Service
So we're doing this again...
And so we come to another turning of the wheel of time, and it is time to announce, once more, the creation of Britain’s FBI.
What I want from a National Data Library
Lots of robust, sustainable open-data, and vibe-coding
I built a tool to fix my local police engagement problem in an hour of café coding. But that was only possible because of a fading public data API from years ago...let's revive the open data dream with the National Data Library!
AI at Play - Lessons from a silly benchmark
Earlier this year, Anthropic treated us to the delightful Claude Plays Pokemon: a wonderful way of watching some poor AI model get lost forever in Mt Moon, fighting a swarm…
Quarto comments, by the open social web
All the comments, all the time, on all the networks? Sure, we can do that.
My 2025 resolution: share more crappy fiddles (and unfinished research)
...and why you should too
I’m a big believer of “working in the open” - open-source, open science, open communication, that sort of thing. It’s not just that I think this stuff is ethnical or right (t…
Weeknote - 2024-W35
Reflections on post-holiday productivity slumps and UK's AI education gaps.
Exploring tools for converting scientific papers to HTML and AI-powered
front-end development. Thoughts on social media disinformation, crime
perception, and the allure of new note-taking apps. Plus, intriguing reads
on slow productivity and necromancers in space.
Weeknote - 2024-W34 (2024-08-19)
On productivity during holidays, sharing thoughts on Copbot Online,
FastHTML, and a new Obsidian plugin. Plus, interesting reads on Llama 3 models
and the challenges of "deliverism" in government.
Copbot Online: bias and variance in AI perceptions of risk
I built a website to look at whether language models could help articulate (or maybe even predict) police risk assessments for missing people. The conclusion? Eeeeh. Maybe. Ish?
Quarto Comments, Powered by the Fediverse
I built a Quarto extension!
I built an extension, and now my blog has fancy comments (and yours can too!)
Why I'm Contributing to OpenStreetMap for Christmas...
...and maybe you should too!
In the eternal words of The Thick of it, everything that happens in the world has to happen somewhere. It sounds obvious, but so many good pieces of research or analysis…
Teaching OpenAI to assess risk, with CopBot!
Looking for the new Copbot Online? Read about it here!
What's Happened to Burglary, and does Attending Help?
It’s nearly 2023, and in the true spirit of Christmas, I’ve used the cheese/wine filled nowhere time between the holidays to pick up an analytical side-project that’s been…
I've migrated to Quarto!
I've migrated to a fancy new technical writing platform, and I'm a little in love with it
Hackers aren't excited about AI (and other bits from CCCamp23)
So I’m sitting in Berlin airport, waiting to get home from Chaos Communication Camp 2023. It was, to put it bluntly, bloody fabulous… and my god, I’ve missed hanging out…
Learning GIS in Python - Robbery and Police Searches in Space
Using Python GIS libraries to explore the spatial distribution of robberies in London
Learning R - Exploring the COVID Crime Effect in London
I use public London crime data on robbery and burglary to examine where this COVID crime shift was strongest, and whether any specific drivers or correlates can be identified.
LAPD Call Prediction for Fun (and Prophet)
Predicting LAPD police call demand using the Prophet time series forecasting library
Twitter Sentiment Analysis with FastAI
Using open source twitter data to explore sentiment in public debate
No matching items