Hackyard for Business Case study
26 teams in two hours. A fine-tuned model in one day.
01 The partner
Adaption turns rough data into training-ready datasets and runs training experiments.
02 What we ran
A Best Use of Adaption track at Hackyard Build 2026, a 24-hour hackathon at IIT Guwahati. Builders put Adaption to work on real problems.
03 The numbers
- 26 teams
- Signed up to build with Adaption, in two hours.
- 71 builders
- Across those 26 teams.
- 1 day to a fine-tuned model
- From 134 examples, for the winning team.
04 The winner
Best use of Adaption
Team Vana · SecondKey
SecondKey finds AI-built apps that leak one user’s private data to another. The team needed a small AI judge it could run on its own machine, so scanned data never leaves it.
- 134hand-labelled examples of leaked tables
- 1,030training rows built with Adaptive Data
- 3BLlama 3.2 LoRA fine-tune, two AutoScientist runs
Adaption reviewed the team’s runs before naming them the winner. SecondKey also took 1st place overall at Build 2026.
Neel Jogani · Anuj Singh · Vinit Chaudhari · Ananya Singh
Also built on Adaption
- TriageLens Support tickets into model-ready data, through the Python SDK.
- TIUL Cleaner sensor data for coal-mine maintenance.
- Sushruth Medical reports explained in Indian languages.
Shyamol and the Hackyard team made our Build 2026 partnership smooth and quick, and they put Adaption in front of a talented group of builders. That’s why we came back for a second event with the Hackyard Dataset Challenge.
Same playbook. Your product.
Pilots from $1,000. You only pay for users your own dashboard confirms.
Run a challenge