The Strange Data Project Kicks Off at Vibeworks

01.A Competition About Data, Not Prompts

Last Friday, Ben Barrett and Thomas Von Teichman from Nomad AI kicked off the Strange Data Project at Vibeworks Dublin. Anthony Fildes and I hosted the space and helped people get set up with their projects.

The premise is straightforward: find data that language models can't currently access—whether it's locked behind paywalls, exists in non-text formats, or simply hasn't been collected before—then transform it into something usable and show measurable improvement on a real task. No expensive retraining allowed.

It's a good challenge because it forces you to think about data problems rather than just prompting better, which is where a lot of the real value in AI work actually sits.

What counts as strange data?

Strange data comes in four broad kinds, and every team's project starts with one of them:

• Access-locked data sitting behind paywalls, login walls or enterprise systems.
• Non-text formats such as images, audio, video, sensor data and CAD files.
• Uncollected data: domain expertise that has never been written down.
• Real-time data from live feeds, streams and other sources that keep changing.

The key constraint is that nobody retrains a model. The competition rewards clever data transformation and representation rather than compute, so every team has to show a measurable improvement using existing models and techniques.

02.How It Works

Teams of 1–5 people have 16 days to build their submissions. The competition runs from January 31st through to Demo Day on February 15th, back at Vibeworks. Entry is free, through Nomad AI's registration at nomadai.ie.

03.€5,000 Prize Pool

At least one team member needs to attend Demo Day in person on February 15th to claim prizes. Prizes are distributed via bank transfer within 14 days, and plagiarised code that isn't open source means immediate disqualification.

04.The Kickoff

The morning started with Ben and Thomas walking through the competition structure—scope of work, deliverables, evaluation criteria, and timeline. Then we shifted into a working session where teams formed up and started scoping their projects.

Anthony and I helped people figure out their approaches. A lot of the early conversations were about identifying genuinely novel data sources—things that aren't sitting in a convenient API somewhere. Irish cultural archives, environmental sensor data, domain-specific records that haven't been digitised properly. The kind of stuff that's valuable precisely because it's hard to get at.

The working session ran until the afternoon, with teams getting their initial plans together and figuring out what they'd need to build over the next two weeks.

05.Demo Day: February 15th

Teams now have until February 15th to complete their submissions. Demo Day will be back at Vibeworks where teams present their data pipelines, show baseline vs. enhanced comparisons, and make their case to the judges.

If you're interested in what comes out of it, keep an eye on Nomad AI's competition page for updates.

FAQFrequently Asked Questions