Skip to content
Registration openOnboarding Nov 18

Adaption Dataset Challenge: Adapt the data. Then prove it.

Build an instruction-tuning dataset for an underserved language or domain, train a model on it and measure the change.

Starts, submissions open
Submissions close
Judging
Dec 7 to 13
A tower of glass data tiles, neatly stacked and iridescent at the top and broken into loose voxels at the base.

Buildaninstruction-tuningdatasetforanassistantinanunderservedlanguageordomain.ImproveitwithAdaption,trainamodelwithAutoScientist,thenshowwhatchanged.

How the challenge works

  1. A pearl-white voxel cube that breaks into loose pixels on one side and builds back up on the other.

    Pick a use case
    that matters

    Choose an assistant for an underserved language or domain, the people it serves and how it should behave. The use case is yours to choose.

  2. A chrome ribbon running through a clear glass lattice.

    Build the data
    with Adaption

    Start from a seed dataset and create or improve your instruction-tuning data with Invent a Dataset and Adaptive Data.

  3. Two crystal blocks side by side, one cloudy and cracked with flecks of pixels, the other clear.

    Train it
    and prove it

    Train a model with AutoScientist, evaluate it with the approach agreed with Adaption, and show the improvement.

Access, resources and rewards

  • 3 weeks

    to build, from Nov 16 to the Dec 6 deadline.

  • +$1,000

    in Adaption platform credits for each member of the top three teams.

  • $100

    in cash for the first-place team, plus 2 VIP swag boxes.

Adaption will provide participants with the platform access and resources needed to build datasets and train models during the challenge.

Start with Adaption’s documentation and the welcome deck and quickstart. Onboarding with Adaption on Nov 18 covers account setup, dataset preparation, Invent a Dataset, Adaptive Data and AutoScientist.

Credits have no cash value. You can’t transfer them, resell them or exchange them for money.

Prizes for the top three teams

1st place

+$1,000in credits, per member

  • $100 cash for the team
  • 2 VIP swag boxes

2nd place

+$1,000in credits, per member

  • Swag boxes

3rd place

+$1,000in credits, per member

  • Swag boxes

Each member of the top three teams receives an additional $1,000 in Adaption platform credits. Only submissions that use Adaption in their tool stack can win the cash prize. All amounts are in US dollars.

Swag ships once results are final. We’ll ask winners for a name, shipping address and phone number, and use them only to ship the boxes.

Dates and deadlines

  1. NowRegistration openSign up and team up before the start.
  2. Nov 16Challenge starts, submissions openFully online, from anywhere.
  3. Nov 17Adaption account verification opensOne PDF per team, on Unstop, by Dec 6.
  4. Nov 18Participant onboarding with Adaption9:00 AM EST / 7:30 PM IST
  5. Dec 6Submissions and account verification close11:59 PM IST
  6. Dec 7 to 13Verification and judgingResults date: to be announced.

You get three weeks to build, and judging runs Dec 7 to 13. These dates are planned and may change.

Participant onboarding with Adaption is on , 9:00 AM EST / 7:30 PM IST. Submissions close Dec 6 at 11:59 PM IST.

How to submit your project

  • A dataset created or improved on Adaption's platform.
  • A model trained with AutoScientist.
  • A description of the use case, the intended users and the assistant's behaviour.
  • How you used Adaption, including the features, settings and iterations.
  • Evaluation results showing how the model performs, with a baseline comparison where agreed.
  • A README or write-up covering your approach, data source and licence, results, limitations and team members.
  • A demo of the assistant and the work you completed.

Submit once, on one platform only: Unstop, or your team page in the Hackyard app at app.hackyard.org/e/adaption-labs. Share one link to your work that contains everything listed.

How judging works

Hackyard runs judging end to end. Adaption verifies signups and dataset runs, and a judge from Adaption’s team sits on the final round.

  • Dataset quality. The quality, relevance and suitability of the dataset for your use case.
  • Measurable model improvement. Evidence that the model you trained with AutoScientist improves on the intended task, using an evaluation method agreed with Adaption.
  • Relevance of the use case. How important and clear the problem is, who the assistant is for, and its practical value.
  • Clarity of the write-up. How clearly you explain your decisions, workflow, results and limitations.

Frequently asked questions

What is the Adaption Dataset Challenge?

The Adaption Dataset Challenge is a fully online challenge run by Hackyard with Adaption Labs. Teams build an instruction-tuning dataset for an AI assistant serving an underserved language or domain, improve it with Adaption, train a model with AutoScientist and show the result. It runs from Nov 16 to Dec 6, 2026, and registration is open now.

Is registration open?

Yes. Registration for the Adaption Dataset Challenge is open now, and the challenge starts on Nov 16, 2026. Participant onboarding with Adaption is on Nov 18, 2026, 9:00 AM EST / 7:30 PM IST. Submissions close on Dec 6, 2026, 11:59 PM IST, and verification and judging run from Dec 7 to 13.

Can I register on Unstop?

Yes. You can register for the Adaption Dataset Challenge on Unstop instead of this site. Registration there runs from Nov 7 to Dec 6, 2026. Register on one platform only, since either one is enough to take part. The Unstop listing is at https://unstop.com/hackathons/adaption-dataset-challenge-hackyard-technologies-pvt-limited-1799281.

Do I get a certificate?

If you registered on Unstop, yes. Unstop issues participation certificates to people who registered for the Adaption Dataset Challenge on Unstop, and winner certificates for the teams placing 1st to 3rd. Certificates come from Unstop, so they go only to people who registered there. Unstop registration runs from Nov 7 to Dec 6, 2026.

Is it online?

Yes. The Adaption Dataset Challenge runs fully online, so there is no venue to travel to and you can take part from anywhere, subject to applicable laws and Adaption's platform terms. The challenge runs from Nov 16 to Dec 6, 2026, with verification and judging from Dec 7 to 13.

What do I build?

You build an instruction-tuning dataset for an AI assistant serving an underserved language or domain. Pick a practical use case, prepare a seed dataset, create or improve it with Invent a Dataset or Adaptive Data, train a model with AutoScientist and evaluate it. A question-answering assistant for an underserved language is one example; the use case is yours to choose.

What do I get access to?

Adaption will provide participants with the platform access and resources needed to build datasets and train models during the challenge. Adaption's documentation at docs.adaptionlabs.ai and the onboarding session on Nov 18, 2026, 9:00 AM EST / 7:30 PM IST cover account setup, dataset preparation, Invent a Dataset, Adaptive Data and AutoScientist.

How do I set up my Adaption account?

Create your account at adaptionlabs.ai/app with the same email you used to register for the challenge, because Hackyard shares it with Adaption for account access, credits and participation verification. If you already have an Adaption account with that email, use it. The onboarding session on Nov 18, 2026, 9:00 AM EST / 7:30 PM IST covers setup.

What is the account verification step?

Every team member takes a screenshot of their Adaption account registration, and the team leader uploads them as one PDF to the Account Verification stage on Unstop by Dec 6, 2026, 11:59 PM IST. Solo participants upload their own. It is a supplementary check, separate from your project submission, which it doesn't replace.

What data can I use?

Public data or a dataset your team creates, as long as you have the right to use and adapt it and you name its source and licence. Use instruction and response pairs or chat-style data in a format Adaption supports. Personal, confidential or unauthorised data, and unsafe or illegal content, are not allowed.

Do I have to use Adaption?

Yes. Your dataset is created or improved on Adaption's platform and your model is trained with AutoScientist. Only submissions that use Adaption as part of their tool stack are eligible for the $100 first-place cash prize. If no eligible project is submitted, the cash prize isn't awarded.

Can I enter solo, and who is eligible?

Participants must be at least 18 years old to use Adaption. The challenge is fully online, so you can take part from anywhere, subject to applicable laws and Adaption's platform terms. Teams have 1 to 4 members, so you can also enter solo. Questions about eligibility can go to [email protected].

When can I submit?

Submissions open on Nov 16, 2026, the day the challenge starts, and close on Dec 6, 2026, 11:59 PM IST. You can register and team up now, but you can't submit before Nov 16. Submit once, on one platform only: Unstop, or your team page in the Hackyard app at app.hackyard.org/e/adaption-labs. Share one link to your work that contains everything listed.

What do I submit?

Your dataset, details of the model you trained with AutoScientist, your evaluation results, a README or write-up covering your approach, data source and licence, results, limitations and team members, and a demo of the assistant. Submit once, on one platform only: Unstop, or your team page in the Hackyard app at app.hackyard.org/e/adaption-labs. Share one link to your work that contains everything listed.

Where do I get challenge updates?

Hackyard posts participant updates in the challenge WhatsApp group at https://chat.whatsapp.com/DyiXXOZ4ATJ7xVRUpxyKn8?s=cl&p=a&mlu=4&ilr=4. Adaption's documentation is at docs.adaptionlabs.ai, and the welcome deck and quickstart cover the first steps. Questions can also go to [email protected].

How does judging work?

Hackyard runs judging. Adaption verifies signups and dataset runs, and a judge from Adaption's team sits on the final round. Projects are judged on dataset quality, measurable model improvement, relevance of the use case and clarity of the write-up. Verification and judging run from Dec 7 to 13, 2026.

Who owns my project?

You do. Participants keep ownership of the projects they build for the Adaption Dataset Challenge. When you submit, you share your dataset, model details and write-up so Hackyard and Adaption can verify and judge the challenge, and the project itself stays yours.

How is my email used?

Hackyard will share participant email addresses with Adaption for challenge-related account access, credit allocation and participation verification. Further details about data handling will be provided in the final participant rules and applicable privacy notices. Adaption's standard terms and privacy policy apply once you create an account.

Can the dates change?

Yes. The dates are planned and may change. As planned, the challenge starts Nov 16, 2026, onboarding is on Nov 18, account verification and submissions close Dec 6 at 11:59 PM IST, and verification and judging run from Dec 7 to 13. Results date: to be announced.

A chrome grid cube floating in the warm light of a desk lamp.
Registration open

Register now.
We start Nov 16.

It runs online from Nov 16 to Dec 6, 2026, and judging follows from Dec 7 to 13.

Questions? [email protected]