Rapidata has built a platform that gamifies RLHF by distributing human review tasks to 15–20 million users across popular apps like Duolingo and Candy Crush, delivering near real-time feedback for AI models.

Founder Jason Corkill launched Rapidata to tackle the data annotation bottleneck he observed during his ETH Zurich work in robotics and computer vision.

The company markets itself as an infrastructure layer that makes human judgment broadly accessible, enabling taste-based and contextual feedback for generative AI beyond traditional objective labeling.

Industry executives note the value of rapid, real-world feedback across diverse markets and languages, highlighting the breadth of potential impact.

Rapidata recently raised an $8.5 million seed round led by Canaan Partners and IA Ventures, with participation from Acequia Capital and BlueYard, to scale its approach.

Key metrics include a user network of 15–20 million, the ability to process about 1.5 million annotations per hour, and support for roughly 5,500 humans per minute providing live feedback.

Looking ahead, the vision includes real-time, large-scale human input where AI models consult vast populations to optimize design and content, blending silicon with society.

The system turns user attention into data-labeling tasks offered as an alternative to watching ads, enabling millions of users to submit feedback in seconds rather than weeks and accelerating RLHF cycles dramatically.

The platform supports online RLHF by integrating feedback directly into training loops on GPUs, enabling thousands of GPUs to receive real-time human judgments and reducing reward-model hacking risks.