Introduction

Planterary Scale AI Training Data using Prediction Markets.

Reppo's mission is to build a decentralized version of Scale AI on-chain where everyone involved in the generation and monetization of AI training data shares the upside, without intermediaries involved.

Data is generated on Reppo.ai, our flagship product built on top of Reppo protocol and deployed on Base that allows anyone in the world to start collecting and contributing to domain-specific AI training datasets. Data is monetized in a permissionless manner on Reppo.exchange, a seperate platform built on top of Reppo Protocol scheduled for Q1 2026 release. The Reppo protocol is a generalized incentive protocol* that enables market mechanics to crowdsource pretty much anything. We uniquely use it to generate AI training data on-demand by introducing the concept of Stake-Assured Human Feedback - Data labeling and annotation backed by economic stake.

*Anyone can use Reppo Protocol to build products and services on top. We are proving the first $1B product on top, inspired by Hyperliquid. This novel utilization of Reppo Protocol which we discuss extensively below enables a decentralized network of AI training data engines called subnets, where each subnet is a prediction market in itself. Subnets collaborate between each other and act as a on-demand data factory for AI models, agents, robotics, and other large scale AI systems.

Reppo might be thought of as "Bittensor for AI training Data" to understand the vision of the core contributors, although the underlying incentive mechanism, consensus rules, and architecture are entirely different. Current Gaps in AI Training Data

Traditional approach to generating Post-training data, specifically Data Labeling & Annotation, operates on a pay-per-task model that prioritizes speed over quality, leading to:

• Rush work to maximize volume

• No accountability for accuracy

• Shallow binary labels

• Heavy QA overhead

Reppo's approach

Reppo disrupts this model turns the entire process into a prediction market between miners and validators generating high-quality verified AI training data at planetary scale.

"Miners" on the Reppo network are primarily source data (raw data) contributors. This is anyone who has unlabeled or non annotated data like geospatial, robotics, adult content, coding etc.

These miners participate in a network of public and private subnets of their choice and provide proprietary human-generated input data. We have plans to integrate with Worldcoin to differentiate humans from AI agents but we believe AI agents can also mine $REPPO. On the other end, validators i.e. voters, motivated by profit, not salary, lock REPPO to acquire voting power (VeREPPO) and vote on miner published data while labeling + annotating it. Note that the voter has no interest or motivation in being data labelers and annotators. They just end up doing the job because they are motivated by economic incentives. Voters on Reppo Network can be compared to "validators" who redirect network emissions every epoch to miners based on the quality of their work. Reppo is not an L1 or L2 network (yet)*.

&Once we start averaging 50k+ daily transactions onchain, we will consider launching our own privacy-enabled chain as we need this for getting rid of the commit-reveal mechanism and improve UX and latency. We believe this is the right way to do it.

Product -> Users -> Scale -> Own Chain.