NeuroLedger

What it is: NeuroLedger is an institutional-grade, policy-aware DeFi lending protocol. It completely separates dynamic risk policy computation from on-chain execution by bridging Zero-Knowledge proofs, Chainlink CRE, and AI.
The Problem: Traditional DeFi lending is open (exposing borrower positions/identity), static (relying purely on hardcoded LTV math instead of external market context), and reactive (delegating liquidations to blind third-party bot wars). This prevents institutional capital from adopting decentralized credit rails.
How it works: We split the protocol into three layers:

1. Private Eligibility: Borrowers generate an off-chain Groth16 ZK proof binding their amount, nullifier, and root. BorrowGate verifies this on-chain without exposing identity.
2. Off-Chain Policy (CRE): A Chainlink CRE workflow intercepts the BorrowRequested event, reads protocol state, fetches external market sentiment (Fear & Greed index), and prompts a Gemini AI agent for a credit risk score.
3. On-Chain Enforcement: The AI acts as an advisor but is constrained by deterministic protocol guardrails (automatically rejecting if riskScoreBp > 8000). If approved, the CRE writes the decision to the BorrowApprovalRegistry, and the smart contracts enforce the final execution.
   A second cron-driven CRE workflow continuously sweeps the protocol for solvency, computing exact repayment amounts off-chain and triggering partial or full liquidations deterministically on-chain.

NeuroLedger utilizes a modular, three-tiered architecture to strictly separate off-chain policy from on-chain execution:

• Zero-Knowledge Privacy Layer: We implemented Groth16 ZK-SNARKs to create a private "ZK-Pass." This allows users to cryptographically prove their borrowing eligibility on-chain without exposing sensitive identity or financial parameters to the public.
• Off-Chain Policy Engine (Chainlink CRE): Built with TypeScript using the CRE SDK, this acts as the protocol's decentralized brain. We engineered two distinct workflows: an event-driven "Borrow Risk Orchestrator" that intercepts loan requests, and a cron-driven "Solvency Sweep" that autonomously calculates optimal liquidation repay amounts.
  CRE Borrow Workflow: https://github.com/hashimh399/zk-private-nl/blob/main/workflows/zkpass-risk-orchestrator/index.ts
  Liquidation Workflow: https://github.com/hashimh399/zk-private-nl/blob/main/workflows/liquidation-orchestrator/index.ts
• Decentralized AI & Data Integrations: We built a custom Gemini AI risk agent and integrated external market sentiment (Alternative.me Fear & Greed API). Crucially, we utilized CRE's ConsensusAggregationByFields to fetch a median AI risk score across multiple decentralized nodes, preventing a single point of failure before writing the decision.
• On-Chain Execution Layer (Solidity): Deployed on Ethereum Sepolia, our custom smart contract suite (LendingPool, BorrowGate, BorrowApprovalRegistry, and CRE Receivers) serves as the ultimate deterministic enforcer, ensuring no AI decision bypasses hardcoded protocol solvency limits.