scale.fun

Problem Statement

scale.fun connects AI Labs needing labeled training data with a global, crypto-incentivized workforce on mobile.📱 AI Labs upload raw datasets (images, audio, text) and set bounties in cUSD. Workers complete micro-labeling tasks on their phones through MiniPay and get paid instantly. All labeled datasets are stored on Filecoin and sold on our decentralized marketplace.Key features: 🔐 ZK-verified identity via Self Protocol (age/country verification for region-specific data labeling) 💰 Instant cUSD micropayments on Celo for each completed annotation 📦 Filecoin storage for immutable dataset archiving 🛒 Decentralized marketplace where AI Labs can buy/sell verified labeled datasets 🌍 Mobile-first design optimized for MiniPay users in emerging markets like Africa, Latin America, India 🇿🇦🇦🇷🇮🇳This solves the massive bottleneck in AI development: getting high-quality, ethically-sourced training data at scale.

Solution

Built as a Celo MiniApp using Next.js 14, optimized for MiniPay mobile wallets in emerging markets.⚙️ Smart Contracts (Solidity):DataAnnotateEscrow: Manages worker bounties and payouts in cUSDDatasetMarketplace: Handles dataset sales with 15% platform feeDeployed on Celo Mainnet (42220)📦 Filecoin Integration:Synapse SDK for uploading labeled datasets to Filecoin Calibration TestnetFast retrieval API for buyers to download purchased datasetsStores both raw files and annotation JSON with content-addressed CIDsBackground upload queue to handle large dataset batches🔐 Self Protocol (ZK Verification):Zero-knowledge proofs for age/country verification without exposing PIIEnables task gating for location and age-specific data labeling requirementsStored on-chain for trustless verification🛠️ Tech Stack:Frontend: Next.js 14, TailwindCSS, Wagmi/Viem for wallet connectionsBackend: Supabase (PostgreSQL) for user data, annotations, and marketplace listingsStorage: Filecoin Calibration Testnet via Synapse SDKBlockchain: Celo Mainnet (real cUSD payments)💡 Hacky Bits:Mixed testnet/mainnet: Self Protocol on testnet, but marketplace on mainnetBackground Filecoin uploads: Built queued API for 60-90s uploadsMobile-first download: Custom component for MiniPay WebView constraintsGrouped dataset logic: Multiple files share one on_chain_dataset_id but separate CIDs