What happens when a hedgehog story prickles its way into the BBC

Researchers at Cambridge University used AI and a 20-petabyte geospatial model called TESSERA to help identify hedgehogs in images, leading to a BBC News article and a flurry of media appearances. The story highlights the challenges of training large AI models, including jury-rigged storage clusters under desks, and the intense national news cycle that followed.

Silviu Petrovan https://www.cambridgeconservation.org/about/people/dr-silviu-o-petrovan/ and I had quite a week appearing on the BBC website https://www.bbc.co.uk/news/articles/c202g60qrlpo and a dozen radio and TV shows talking about the applications of AI to saving hedgehogs https://anil.recoil.org/ideas/hedgehog-mapping . I thought I'd jot down a "behind the scenes" so other academics get a sense of what to expect when a research story hits the national news cycle. Mostly, make sure you have a clear couple of days to deal with live media appearances with just a few hours notice and don't wear a rumpled shirt to the office Zoe Kleinman from the BBC pinged me after hearing from Vultr https://geotessera.org/blog/2026-03-30-training-and-inference-at-scale about the TESSERA v1 geospatial foundation model we've been developing here. I grabbed Silviu Petrovan https://www.cambridgeconservation.org/about/people/dr-silviu-o-petrovan/ and Kevin Cochrane https://www.linkedin.com/in/kevincochrane from Vultr https://www.vultr.com and we had a brisk 25-minute chat on Teams with Zoe and her producer. This was all very stress-free and just like talking to anyone else, since it wasn't a live segment where every word had to be perfect. Zoe published the article on hedgehogs https://www.bbc.co.uk/news/articles/c202g60qrlpo later that day straight onto the BBC website. I hadn't realised before this just how widely read the BBC site is, as it has over 95% of UK adults https://www.bbc.co.uk/aboutthebbc/documents/ara-2023-24-our-audiences.pdf using it and over half a billion unique visitors per month https://www.bbc.com/mediacentre/2025/bbc-reaches-record-audience-digitally-annual-report-and-accounts . To be able to accurately identify the tiny creatures and other objects in image data, the Tessera system had to be trained on vast amounts of data — with around 20 petabytes, or the equivalent of 10 billion standard digital photos, used to build it. After reaching the limits of the computing power allocated to the university, researchers installed additional processors under their desks to keep the process going. -- Zoe Kleinman, BBC News, May 2026 I chortled when reading that last paragraph about 'under their desks' as I'd mentioned that over dinner to Zoe back at the OpenUK AI Impact Summit roundtable https://anil.recoil.org/notes/path-to-uk-india-ai-summit a few months ago that Amanda Brock https://amandabrock.com/ invited me to. The jury-rigging is, for the record, entirely true, since we've had to cobble together a gigantic storage cluster https://www.tunbury.org/2026/03/27/ceph-expansion/ which we've only recently been able to move offsite https://www.tunbury.org/2026/04/17/cephfs-to-s3/ . The way these radio sessions work is that a researcher gives you a call ahead of the live slot, checks audio tip: don't use Airpods as they sound echo-y, just speak normally into the phone handset instead , and then gives you a time for the live slot when they'll phone over Facetime in my case as I have an iPhone . When the call comes in, your audio is mixed in with the radio show running in the background but muted , and then before you know it you're live on national radio with millions of listeners talking to the presenter They only call a few minutes before, presumably so that you don't get too nervous thinking about it all. After this, the TV crews showed up to Cambridge to film pieces for television. Since it's exam term and College is in a quiet period https://www.pem.cam.ac.uk/current-students/policies-procedures-guidance/college-regulations we couldn't film onsite as we've done in the past https://anil.recoil.org/notes/cacm-docker-cover and so used my office. They show up with a fairly portable set of camera and sound equipment and lugged it up the three floors to my office. The first TV slot https://crank.recoil.org/w/dxXkfLYocHMbZtdkZLAc8R was from BBC Look East, and I grabbed David Coomes https://coomeslab.org and Sadiq Jaffer https://toao.com at short notice from across the road to get some more team voices included. Having my overgrown plant-filled office did seem like quite a nice idea on paper, but I think it came out less well when actually on TV since it's zoomed in and the light isn't even. I also hadn't reckoned on the sneaky 'reverse' shots the cameraman took which exposed the completely disorganised other half of my office away from the nice curated end We had about 30 minutes with the production team as well for this video, since they also doubled as the camera crew. We discussed what shots to take, and they did capture some gorgeous interactive views of the TZE Explorer https://tze.geotessera.org which you can see in the clip above and I love the studio intro which zooms in on the parametric UMAP false colour TESSERA map https://www.cl.cam.ac.uk/~avsm2/cb2-pumap.svg of Cambridgeshire . ITV turned up later with another very patient cameraman to do a "down the line" piece https://crank.recoil.org/w/u34bg5FqQQ7618hN2DpsKe , which was a quite different structure from the previous one. The presenters were all in the recording studio remotely, and I got wired in with an earpiece to listen in to them. The filming then happened as if I was in their studio, with compositing to make it all work. The other thing to know is that the university comms office is actually monitoring the media mentions of the University on an ongoing basis, since Cambridge regularly goes viral with some mention in the news almost every day. The team sent me a weekly recap of all the mentions, including an interesting service called TVEyes https://tveyes.com who they pull clips of every mention across the BBC, ITV, Sky and the regional stations and forward them on. This is the only way I have any idea of knowing what actually went out in several cases since rebroadcasting a story is quite common. And thanks to an email from Rachael Pells, we also appear in the print edition of The Times not online for some reason with a lovely little section. I missed this on the first outing until Julia P.G. Jones https://www.bangor.ac.uk/staff/sens/julia-patricia-gordon-jones-010356/en mentioned her parents were reading https://www.linkedin.com/feed/update/urn:li:activity:7465356682768928769/?dashCommentUrn=urn%3Ali%3Afsd comment%3A%287465380895626022912%2Curn%3Ali%3Aactivity%3A7465356682768928769%29&dashReplyUrn=urn%3Ali%3Afsd comment%3A%287465382045167276033%2Curn%3Ali%3Aactivity%3A7465356682768928769%29 the article The thing I was most worried about, going in, was the science communication standard required for a national audience, especially since climate change messaging https://doi.org/10.1080/1533015X.2026.2628164 is under such threat at the moment. Biodiversity, however, seems to trigger a different and perhaps more instinctual response in people, and there was no one denying that there exists a crisis https://anil.recoil.org/papers/2025-biodiversity-9recs in biodiversity. Instead, much of the online conversation was about differences in opinion about how to solve the land use problem from conflicting solutions like should we cull badgers https://www.badgertrust.org.uk/post/badgers-and-hedgehogs-separating-fact-from-fiction-in-british-native-wildlife to save the hogs . This feels like an area where TESSERA https://anil.recoil.org/projects/tessera , Enki https://anil.recoil.org/projects/enki and Conservation Evidence https://anil.recoil.org/projects/ce could all have a positive impact by making facts about the planet more available, so I feel quite energised about that I was very lucky that my collaborators esp. Silviu Petrovan https://www.cambridgeconservation.org/about/people/dr-silviu-o-petrovan/ in conservation are old hands at this sort of thing, and they kept the focus on the core conservation message which is the thing most of interest to a national audience. I also think I spent a bit too long talking about the innards of TESSERA https://anil.recoil.org/projects/tessera , and need to practise a few snappy lines about what it is at its essence. For example, Sadiq Jaffer https://toao.com came up with the term 'satellite fingerprint' to describe a TESSERA 10m tile, which seems hella snappier than '128-dimensional embedding of Sentinel 1 and 2' when doing an elevator pitch. All in all, I very much enjoyed the experience. It has completely disrupted my already packed work-week, but I must have received a hundred messages from friends who otherwise couldn't care less about my day-to-day research, and it gave the project https://geotessera.org/blog/2026-05-19-tessera-hedgehogs 1 , department https://www.cst.cam.ac.uk/news/tessera-ai-model-helping-protect-hedgehogs-space and College https://www.instagram.com/p/DYnL2pTDkkb/ something to post about that doesn't involve obscure functional programming Thank you Zoe Kleinman for taking the time to run this piece