Metaphor

Demo

About

The problem it solves Institutions wanting to tokenize real-world assets (bonds, invoices, real estate) face three barriers: their portfolio strategies are exposed on public blockchains, coupon/yield payments depend on fragile off-chain servers, and risk analysis is manual and disconnected from on-chain data. Outward solves all three in a single platform — asset managers tokenize RWA with manual KYC whitelisting and role based access control, negotiate privately through confidential vaults where each party only sees what they're authorized to see, automate coupon payments deterministically on-chain with no centralized infrastructure, and get AI-powered risk analysis with stress testing and actionable recommendations that always require human approval before execution. Use of AI tools and agents Outward integrates an AI-powered risk analysis assistant built on 0G Compute for decentralized inference. When a user clicks "Analyze" on a vault, the backend sends the vault's composition (asset types, ratings, allocations, maturities) to 0G Compute, which returns a structured risk report — a composite score from 1 to 100, per-position analysis, stress test simulations (rate hikes, issuer defaults), and actionable rebalancing recommendations. The output is validated against strict schemas (Zod) to ensure consistency. Critically, the AI never executes transactions on its own — every recommendation is surfaced in the dashboard with explicit approve/reject buttons, enforcing a human-in-the-loop workflow. A mock inference mode is available as fallback when 0G is not configured.

How it's made

Challenges we ran into Our InstitutionRegistry contract hit Ethereum's 24KB deploy limit (EIP-170), so we extracted an InstitutionDeployer to handle the heavy logic externally. The backend had an ESM compatibility issue where ABI file loading broke because __dirname doesn't exist in ES modules. Hedera's Schedule Service precompile had limited documentation, so we relied on iterative testing on testnet. And Daml was new to the whole team — Canton's party-scoped visibility model took time to get right.