At Reppo Labs, we are building platforms and protocols to orchestrate a three-sided marketplace to connect DeAI infra providers, developers/creators, and end-users of verifiable ML/AI.
What is verifiable ML/AI?
After spending over 10 months with end-users within Web3 communities, specifically web3 governance, web3 investment firms, as well as DePINs, it has become clear that mere “verifiable inference”, i.e. obtaining a proof that an inference was served by a specific model/agent does not solve the pain of customers who want to know how the model was trained and who built it to truly underwrite risk and showcase decentralization with their community.
Take for example Marta Piekarska, who currently leads Filecoin Incentive Design Labs and has been Reppo’s initial customer
Reppo is building a credibly-neutral permissionless intelligence network that incentivizes collaboration on context-aware intelligence. Reppo builders, organized as pods, pay in native utility tokens to acquire the data and datasets from the Reppo Data Exchange, contributing their context and biases to bring situational awareness to AI models and agents. The initial set of pods are building AI applications for climate, DeFI, insurance, and web3 governance.
The network natively deploys the Reppo Protocol, a set of smart contracts that allows anyone to add ownership attestations, licensing and a monetization scheme to their intelligence.
Reputation models
The most widely adopted use case of reputation models outside web3 are platforms like Yelp, Trustpilot, google reviews etc. Risk/Reputation models also power national-level systems such as credit ratings, rankings for universities, etc. In the world of Web3, risk/reputation models power systems that aid in the validator selection process for staking and liquid staking protocols, replacing centralized committee structures and governance setups, benchmarking SLAs for providers for Decentralized Physical infrastructure Networks(DePINs), and even building verifiable ranking/scoring for creators on various NFT platforms optimized for individual use-cases and needs.
A Reppo model is a risk/reputation model built using web3 primitives of verifiability, ownership, and permissionlessness. Anyone can access the provenance of the data used to build the model and have an on-chain association to the model creator. Reppo models are built by a community of modellers, referred to as Reppo Modelers, who combine various on-chain and off-chain datasets with their own novel heuristics and methods to provide biased frameworks for evaluation & assessment.