Hackathon WinnerETHGlobal EthereumCoinbase Pool Prize
V-WorkForce — Decentralized AI Service Marketplace
ETHGlobal Ethereum winner: autonomous agents with wallets, privacy-preserving legal flows.
Lead Engineer / Architect
AI & AgentsWeb3 & ZKNext.jsCoinbase AgentKitNillion SecretVaultPrivyTypeScript
Agent-controlled wallet executes tasks; SecretVault guards sensitive inputs.
Node
Next.js UI
Node
Agent Hub
Node
Coinbase Wallet
Node
Nillion SecretVault
Node
Ethereum
Next.js UIAgent Hubprompts & intents
Agent HubNillion SecretVaultencrypted docs
Agent HubCoinbase Walletsign & pay
Coinbase WalletEthereumtx settlement
DataControlSecurity
Context
Problem & context
- Legal teams wanted agents that could negotiate, price, and settle blockchain tasks without exposing sensitive contracts.
- Needed privacy-preserving document intake and deterministic on-chain execution for funds held in agent-controlled wallets.
Architecture
Architecture & stack
- Next.js front-end with Framer Motion for interactive agent controls.
- Coinbase AgentKit to give each agent a custodial wallet and on-chain action surface.
- Nillion SecretVault for encrypted legal document processing before agent execution.
- Privy for embedded wallet authentication and seamless onboarding.
Features
Key flows
- Agent-run task proposals that calculate cost and push transactions from delegated wallets.
- Secret document ingestion pipeline guarded by Nillion; agents only see redacted hashes.
- Multi-party escrow flow with milestone releases triggered by agent assertions.
Challenges
Engineering challenges & solutions
- Wallet orchestration for multiple concurrent agent sessions without race conditions.
- Balancing privacy (SecretVault) with verifiability for on-chain receipts.
- Designing UX that reveals agent intent and costs before signing.
Impact
Impact & metrics
- Won Coinbase pool prize at ETHGlobal Ethereum.
- Delivered compliant agent flows for legal documents while keeping signer UX under 2 clicks.
- Demonstrated autonomous wallet execution with human-in-the-loop checkpoints.
Next
What I’d improve next
- Deeper audit trails with zk attestations of agent reasoning.
- AI guardrails for cost caps and automatic risk scoring prior to signing.