Adaption AutoScientist Challenge- $50,000 prize pool

Bring Frontier AI to the World

Powered by Adaption · Run in partnership with HackIndia, India's largest hackathon series

The Problem Worth Solving

Today's AI is expensive, static, and built for the average use case.

A handful of companies ship monolithic models that work well — until your domain, language, or problem doesn't fit the mold. When that happens, you're on your own.

This challenge exists to change that.

What Is This Hackathon?

The AutoScientist Challenge × HackIndia is a global hackathon for builders, researchers, engineers, and domain experts ready to push AI beyond its current limits.

Using AutoScientist by Adaption — the platform that automates the full AI model research and training loop — participants will train and openly release frontier AI models across 10 categories, with the goal of setting new state-of-the-art benchmarks and publishing model weights directly to Hugging Face and Kaggle.

HackIndia brings this challenge to India's global developer community with a dedicated $2,000 prize track for HackIndia-registered participants.

Join WhatsApp channel (necessary): https://chat.whatsapp.com/BAXRPUrf6ht45btgBMPkGO?mode=gi_t

HackIndia Prize Pool: $2,000 USD

🥇 1st Prize: $1,000
🥈 2nd Prize: $600
🥉 3rd Prize: $400

HackIndia track winners are selected from all eligible HackIndia-registered participants competing across any of the 10 categories.

Global Prize Pool — $50,000 USD

10 First Place awards: $4,000 each (one per category)
10 Runner Up awards: $1,000 each (one per category)
30 Honorary Mentions: credits and swagv

HackIndia participants are eligible for both tracks simultaneously.

Timeline

Part 1 (June 22 to July 5)

Categories: Finance, Healthcare, Language, Legal, Marketing
Winners announced: July 13

Part 2 (July 6 to August 3)

Categories: Science, Agriculture, Data Visualization, Math and Code, All Other Domains

Winners announced: August 10

A separate submission form is live for each part. Your category is selected at submission. You may enter Part 1, Part 2, or both.

Exactly What You Need to Do

Step 1 — Sign up
Go to adaptionlabs.ai and create your account. Your 1,000 free platform credits activate at hackathon start.

Step 2 — Pick your category
Choose one of the 10 tracks above. This is the domain your model will be trained and evaluated in.

Step 3 — Build your dataset using Adaptive Data
Use Adaption's Adaptive Data platform to ingest, adapt, and evaluate your training data. The platform supports 242 languages and multimodal inputs.

Step 4 — Train your model using AutoScientist
Run the AutoScientist training loop. It co-optimizes your data and training recipe end-to-end until your model converges on your goal. Free compute is provided throughout the competition.

Step 5 — Beat the baseline
Your model must show a measurable percentage improvement over the baseline model on Adaption's held-out test set for your category. No improvement, no prize eligibility.

Step 6 — Release everything openly
Publish your trained model weights and adapted dataset publicly on both Hugging Face and Kaggle. Include a clear model card and documentation explaining what you built and how.

Step 7 — Post about it
Share your project on LinkedIn and X. Tag @adaption_ai on X and Adaption on LinkedIn. Bonus points for shipping a live demo.

Step 8 — Submit
Part 1 submissions close July 5. Submit here: Part 1 Submission Form
Part 2 submission form will be shared separately before July 6.

How You Will Be Judged

Measurable performance improvement over the baseline model
Quality and originality of your adapted training dataset
Real-world impact and applicability of the domain you chose
Depth of AutoScientist usage in your training pipeline
Open release quality — model card, documentation, reproducibility

What You Get Just for Participating

1,000 Adaption platform credits for data adaptation and model training
Free compute access via AutoScientist for the full competition duration
Access to Adaption's Discord to find teammates and collaborators
Direct office hours and sessions with Adaption's research team
Winning models featured across Adaption's research and community channels

Join the Discord now: discord.gg/THQuQhN7C9 — find the #autoscient-challenge channel

Who Should Join

ML engineers who want to push model performance in a specialized domain
Researchers looking to publish open source model contributions
Students building their first serious AI project
Domain experts in finance, healthcare, law, agriculture, or science who've hit the ceiling with off-the-shelf models
Founders who need AI that actually fits their use case

If you have ever thought "this model would work if it just understood my domain better" — this hackathon was built for you.

About Adaption

Adaption builds adaptive AI infrastructure. AutoScientist automates the full AI model research and training loop, co-optimizing data and training recipes end-to-end so your model converges on your objective — not the average use case.

Adaptive Data, Adaption's real-world training data optimization platform, supports 242 languages and multimodal inputs and is used by researchers and builders worldwide.

adaptionlabs.ai · @adaption_ai