Stockholm · 10–17 November 2026
AI for
Meta-Science.
A seven-day hackathon.
Rethink how science is organised, evaluated, and rewarded. Build the tools to make it happen.
01 / The idea
Science is accelerating.
Let’s help it keep up.
AI is transforming research, from generating hypotheses to running experiments and making discoveries. As scientific output grows, our ability to evaluate, validate, and distinguish meaningful discoveries from noise needs to grow with it.
This hackathon brings together Stockholm AI, the AI for Meta-Science NeurIPS Workshop, and Sweden AI Factory to explore better ways of doing science.
02 / The challenges
Three tasks. Real research problems.
The brief outlines three candidate tasks, from the first research question to the evidence behind a published claim. Open a task to explore what to build.
01Ideation stress-test toolIs this question worth asking?
Researchers can spend months on a question that is vague, impossible to answer with available data, or already answered by prior work. These problems are often caught late. Build an early check on whether a question is precise, testable, and worth the resources.
What to build
A tool that takes one research question and:
- Extracts and restates its key terms, flagging ambiguity or inconsistent use in the literature.
- Surfaces the question’s implicit assumptions.
- Identifies the evidence or data needed to answer it, and flags evidence that is unlikely to be obtainable.
- Recommends whether to proceed, narrow, revise, or stop, with reasoning.
What success looks like
A vague or overly broad question gets a specific, actionable critique. For example, identify the three different meanings of “engagement” in the literature, rather than simply suggesting that the researcher clarify their terms.
02Provenance-preserving lab notebookKeep the research trail intact.
Good research leaves a trail: why a method was chosen, what was abandoned, and which code produced which result. When AI tools rapidly iterate on research tasks, that trail can disappear, eroding accountability and reproducibility.
What to build
A lightweight logging or wrapper tool around a single AI-assisted research task, such as data cleaning, running experiments, or iterating on a model. Automatically capture:
- Each proposed plan or hypothesis and the reasoning behind it.
- What was actually run, including code, parameters, and data versions.
- What failed or contradicted expectations, and how the plan changed in response.
- A timeline a human can follow to reconstruct the actual decision path.
What success looks like
A reviewer who wasn’t present can inspect the log and answer “Why did they change approach at step 4?” without asking the researcher.
03Claim-to-evidence linkerFollow the claim. Find the evidence.
Reviewers rarely have time to trace every claim to its supporting evidence. Polished writing can make a paper feel convincing without making its claims well supported. Build a way to inspect those links directly.
What to build
A tool that takes one paper or preprint, identifies its major claims, and links each to its exact supporting figure, table, results paragraph, or code path. Flag:
- Claims with no traceable supporting evidence in the document.
- Evidence that is weaker than the claim’s stated confidence, such as a strong conclusion backed only by a single small-sample experiment.
- What was checked and what remains unverified.
The output
An inspectable map of claim → evidence → confidence or caveat, so readers can assess each link individually.
03 / Your calendar
Seven days
to build.
All times are Stockholm time (CET).
All dates are in 2026.
- 17:30–20:00
Kick-off in Stockholm
Meet at RISE KTH for an introduction to the challenges and competition details. Food and drinks provided.
Drottning Kristinas väg 61 - 18:00
Submission deadline
Time to share what you’ve built.
Winner announced
The winning team is revealed.
NeurIPS 2026 · Paris
The winning team takes their work to the AI for Meta-Science workshop.
04 / The prize
From Stockholm to Paris.
Your work.
The NeurIPS stage.
The winning team receives an all-expenses-paid trip to NeurIPS 2026 in Paris, 9–13 December, plus a poster slot and a dedicated oral presentation at the AI for Meta-Science workshop.
The hackathon and winning prize are fully sponsored by Sweden AI Factory.
Start with the kick-off
See you in Stockholm.
10 November · 17:30 · RISE KTH
The kick-off is currently full. Join the waitlist on Luma.