Why this one matters
The same operation, by five different surgeons, can produce five different outcomes.
That variability is one of the largest causes of poor outcomes in surgery anywhere in the world. XRlabs exists to fix it, and the way you fix it is computer vision.
Their founder trained as a neurosurgeon and walked into the operating theatre to find technology that had barely moved in twenty years. Surgical navigation with the same underlying tech it had two decades ago, a better interface bolted on top. Multiple screens. Models trained only on the robots that made them.
So he left and started building a surgical foundation model: perception that has seen millions of hours of operating, trained from the surgeon's point of view rather than a machine's. Get that right and you can put a copilot into theatres worldwide, and start closing the gap on the hundreds of millions of operations that never happen at all.
Almost every hard problem they have turns out to be a computer vision problem. There are very few people in the UK or Europe working on them.
That is why this day exists. For one Saturday, their data, their stack and their engineering team are open to twenty people who want to try.
Who is behind it
Leaders in industry.
Real-time AI perception for live surgery, founded by a practising neurosurgeon and running under some of the hardest latency constraints in the industry. Their office. Their data. Their engineers building beside you all day.
Named one of TIME's Best Inventions of 2025, and used to rehearse the separation of conjoined twins. In March 2026, the first use of physical AI in cranial neurosurgery, powered by NVIDIA. The challenge you'll work on is the real problem behind it.
xrlabs.ai · Ali Haddad on LinkedIn · TIME Best Inventions 2025 · First use of physical AI in cranial neurosurgery
Cloud compute for the day: $100 in NVIDIA Brev credits for every participant, so you choose your own GPU.
The deep-tech fund backing the hackathon, in the room and on the day.
The University of Oxford's student-led venture fund, backing the hackathon alongside Playground.
4,500 founders, builders and researchers across Oxford, Cambridge and Imperial. 237 startups through our Challenge last year; finalists have raised over £45m.
Why XRlabs
What else makes them impressive.
- TIME Best Inventions 2025: their XRlabs SIM turns ordinary scans into 3D environments where surgeons rehearse operations
- Conjoined twins: surgeons used it to plan and practise a 14-hour separation of conjoined twins, with Great Ormond Street's Professor Owase Jeelani
- NHS: first installation at an Imperial NHS hospital, where surgeons reported 88% higher confidence after training on it
- Ali: practising neurosurgeon (trained at Imperial NHS), co-founded GoodSAM, the emergency response app used with ambulance services across the UK, and holds a patent in AR surgical navigation
- The why: 143 million surgeries a year go unmet worldwide (Eka Ventures)
The task
Real-time tool differentiation, on surgical video.
You will train and deploy a model that tells surgical instruments apart in live footage, on NVIDIA cloud compute. You build on XRlabs' synthetic surgical video, with a live leaderboard all day and an XRlabs baseline model on it from minute zero. The top five then have their models tested on real surgical video. The specifics land at 10:20 on the 24th, but you know now exactly what you are training for.
Judged on what works
Performance against a hidden test set, false positives and false negatives, with weight for efficiency inside the per-frame budget, and whether your work runs again tomorrow. The top five also present.
Built solo
Twenty individuals, not teams. XRlabs engineers build alongside you, not over your shoulder. That is the part you will remember.
Prepare if you want
Warm up on any open dataset beforehand and bring it with you. A head start on your tooling is fair game. The problem itself lands on the day.
The day
Saturday 24 October
Over six hours of build time. Brunch on arrival, pizza at three, drinks once the presentations are done. All on us.
Where: XRlabs' offices at the London Institute for Healthcare Engineering, 100 Lambeth Palace Road, London SE1 7AR. A lecture theatre and a balcony overlooking Parliament.
Prizes
What you can walk out with.
Handed over that night.
Details to be announced.
Every participant, win or not.
Interested or curious about working in surgical AI? Let us know and we'll try to connect you to people in the team.
Who this is for
Computer vision buffs, wherever you currently sit.
Students and researchers
Oxford, Cambridge, Imperial and beyond. Maths, physics, engineering, stats, life sciences, the department matters far less than the work.
Devs and builders
Already at a lab or a startup, or building on your own. No qualifying task, no homework, and applying takes five minutes.
Anyone who has shipped
A repo, a paper, a competition placing, a thing you built at 2am. Show us, and that is the application.
Apply
Five minutes, and you're in the pile.
We review as applications arrive and send invitations continuously. The first people will start hearing back by Monday 12th October.
Apply
On Luma: your LinkedIn, your GitHub, your university or company, and one thing you've built that you're really proud of.
We read it
Screened by Oxbridge AIx and the XRlabs engineering team together.
You get an answer
The first people hear back by Monday 12th October. If you're in, you sign the NDA and data licence before your place is confirmed. Twenty places, and we hold a handful back for late arrivals worth making room for.
Questions
Before you ask.
Do I need a medical or surgical background?
No. This is a computer vision problem that happens to live in an operating theatre. Everything you need to know about the surgery is explained on the day, and XRlabs engineers are in the room with you throughout.
Do I need to bring a team?
No. Entry is individual: twenty people, each building their own model. Invite friends to apply too!
Do I need my own hardware?
Bring a laptop and a charger. Compute comes from NVIDIA: every participant gets $100 in Brev credits to choose their own GPU.
Can I work on it beforehand?
Yes, and we would rather you did. Warm up on any open dataset and bring your scaffolding with you. The task itself uses XRlabs' data and is revealed on the 24th, so preparation buys a head start on tooling, not on the answer.
Who owns what I build?
You own your code. Before you're invited, you sign an NDA and a data licence: models trained on XRlabs data aren't published or commercialised, and are deleted after the event.
I'm not at Oxford, Cambridge or Imperial.
Apply anyway. The room is picked on ability, and previous winners have come from outside those three. You might meet your future co-founder or bestie!
Is there a cost?
None. Food, snacks and drinks are on us all day.
Run by
×
×
OXFORDSEED FUND
×
×