KGeN presents AI training data collection architecture built on a 53 million-user base

Decentralized data infrastructure company KGeN on the 2nd unveiled an AI training data collection architecture that leverages a user network of more than 53 million.

KGeN is securing data generated in real-world settings through a user base distributed globally, including India, Southeast Asia, Brazil and the Middle East. The core is collecting data that reflects users’ actual intent and behavior—not simple click data.

For data validation, it applied its proprietary "POGE (Proof of Gamified Engagement)" framework. Through this, it verifies users’ identities and activity histories and says it will build a more advanced dataset that excludes bogus data.

It also presented specific examples of dataset construction. KGeN has secured about 20,000 hours of first-person video data capturing daily activities at home.

KGeN emphasized, "Because the data was collected from real life without staging, it reflects the diversity of environments and behaviors as-is," adding, "We operate in a way that strictly excludes personal data and sensitive information during the collection process."

It added, "The quality and diversity of AI training data are becoming key factors that determine model performance," and said, "We plan to continue expanding a sophisticated data architecture generated in real-world environments based on our global network."