About Reppo

Planterary Scale AI Training Data through Prediction Markets.

Reppo is a decentralized blockchain network designed to generate AI training data at scale. The network leverages novel prediction market mechanisms to align economic incentives with data-annotation and high-quality human-feedback on crowdsourced primary/unprocessed data.

The network coordinates data owners, data labelers, data annotators and data consumers to interact, validate and get rewarded for their efforts.

Reppo can thought of as "Bittensor for AI training Data" for a broad strokes understanding of the vision, although the underlying incentive mechanism, consensus rules, and architecture are entirely different. The "miners" on Reppo network are primarily source data (raw data) contributors who participate in subnets of their choice and provide creative and proprietary human-generated input data. On the other end, data labelers and annotators lock REPPO to acquire voting power (VeREPPO) and vote on miner generated data while labeling + annotating it. They can be compared to "validators" on the network who redirect network emissions to miners based on the quality of their work.

Reppo Network is made up of public and private subnets, each of which a specialized mini-network powered by Reppo's incentive framework and focuses on a specific type of AI training data through specific tasks and domains. Different subnets leverage different collection methods and Reppo's platform provides the necessary enablement tooling.

A public subnet allows permissionless contribution from miners and validators (content guardrails permitting) whereas private subnets are owned and operated by individuals. enterprises, and small teams working with proprietary source data and gated access to voters i.e. data annotators to have more control on quality.

Today, Reppo does not operate its own L1 or L2 blockchain. It is a generalized Data-AI coordination layer to crowdsource on-demand human feedback at planetary scale.

---

Our flagship product is a self-serve platform, deployed on Base, that allows anyone to start curating domain-specific AI training data. The public subnet is currrently maintained by Reppo Foundation, where interaction between content creators/publishers and voters is faciliated to gather curated human feedback and preference data. We are also collecting evals metrics on underlying model usage information + benchmarking the various AI tools/platforms being used by miners on Reppo.

We leverage verification technologies such as Project Vail and Lunal to provide provenance to voters on the AI content being published. Additionally, we plan to build an AVS to further quality of source data for data labelers and annotators in Reppo V2.

Enterprises, startups, and even individuals can permissionlessly spin up new subnets and start collecting curated human feedback on crowdsourced or proprietary content, generating high quality RLHF and DPO training data on-demand for AI models and applications such as self-driving cars, mapping, AR/VR, robotics.

Our mission is to build an open and permissionless network where anyone can contribute to training datasets, provide high signal feedback through their domain expertise, and contribute to making the next gen AI models and systems human aligned.