I've taken a Eurovision dataset from Kaggle and analysed it to create this "scrollytelling" article. It uses mostly Scrollama and D3, and includes interactive charts at the end. Methodology and code are included in the article.
In 2020, I joined the NHS AI Lab as its Deputy Director, a £250M UK Government-backed initiative to develop and deploy safe, ethical AI across the UK healthcare system, and one of the first AI incubators ever set up in a public sector context.
As part of it, I ran the AI Skunkworks & Deployment programme, aiming to trial-and-launch AI-driven services.
While I often joke that joining a healthcare organisation in the midst of a global pandemic was a debatable life choice, that experience gave me a solid exposure to what it actually means to try to adopt AI.
I now help run HM Revenues & Customs' Chief AI Office.
I have a never ending list of "small tools" for personal use and decided to try Claude Code to see which ones I have clear ideas about, as I find that there's a good correlation between speed and quality of development with coding LLMs and the ability to capture requirements (in a CLAUDE.md, for example). The latest thing I've created is a tool for a photographer friend, which gives the ability to make equal crops for portrait photos: http://facecrop.puntofisso.net/
1) I'm not 100% sure what to do when the ratio can't be complied with because the original photo has a certain shape, but I'm definitely thinking about it; in general, how to get as much as a consistent crop is the main reason for FaceCrop;
2) trying to keep the UX as simple as possible, so I assume (in this MVP) that the user wants to crop pictures at the same size/quality of the original, and do any further edits elsewhere... but it does make sense and it's probably something not that difficult to achieve with a sensible interface (it also reminds me of https://squoosh.app/)
I have a couple of things that I have mentioned at recent job interviews and helped me get the job, but in a sideways manner. Still make me smile because I essentially did them to learn something and they kinda acquired a life of their own.
Long story short: I had a bit of a fixation with political data wrangling.
This got me two really odd personal successes (excuse the slightly blowing of my own trumpet here, for story's sake): an app [1] that takes UK Parliament debate transcripts and makes an interactive n-gram analysis, similar to Google Books N-gram viewer, which was used by Robert Peston's political show on national TV and the press in the UK (e.g. on the Financial Times [2] and the Sunday Times [3]); then I did a quasi-viral blog post that used code to calculate the average face of a British MP [4], which got me a few contracts, including one with the BBC for the same thing in the US Congress [5]
When I say sideways, what I mean is that the interesting thing is that the jobs I got when using these as examples were not hands-on data wrangling jobs (in fact, they are terribly dirty pieces of code, but that's another story).
What they got me is two things: from a technical perspective, the ability to see an end-to-end process to create a product, the running of a service no matter how small for a decade, the use of cutting-edge technology; from a broader point of view, they were great to show me catching the zeitgeist, seeing stories in data, engaging with national media. Both were incredibly "catchy" stories to tell during an interview, and even when challenged (my recent employers being in the public sector) they allowed me to explain myself and my journey.
So, in summary, I love how these two one-day hacks turned into great interview stories, beyond the very minor direct income that they got me.
Aside from the ability to blow my own trumpet a little, the broader applicable lesson here is that by working on something you have a passion for, no matter how geeky it might be, you can build something simple and not necessarily super tidied up, that will however be a good prompt to discuss both your technical and non-technical skills.
I've coached a few candidates for interviews in the intervening years, especially people in tech roles, and it strikes me how often they play down their own side project, which are sometimes much way technically better than mine and with some pretty good stories around the initial motivation and use examples.
I had a very similar experience with TripAdvisor and a hotel in Vietnam, which claimed my review was false. The really crazy part of this was that my review was positive, 4/5 stars. But they wanted 5 stars and "befriended" me on Instagram to ensure I did that. Even with all this proved and screenshotted, TripAdvisor didn't do much until I went public on social media (thread: https://twitter.com/puntofisso/status/955726203662536704) and some highly visible users piled in.
I can't trust any reviews after this. The major issue is that the review system is flawed. One odd bad review should not affect the overall standing of a business, so even assuming a review was flawed they should leave it on; if they don't, it means that the whole review framework is wrong somehow because statistically gives power where it shouldn't.