Helm — AI × RWA on Mantle

Dynamic yield strategies and automated risk management for RWA assets (USDY, mETH) built on Mantle's RWA infrastructure.

Helm is an AI portfolio copilot for real-world-asset and liquid-staking yield on Mantle. It reads live on-chain and market data, computes a risk-budgeted target allocation across real Mantle yield venues, and runs automated risk management (peg health, concentration, liquidity, idle capital) on the book — with an AI engine (via OpenRouter) generating the strategy thesis and risk narrative.

Built for the Mantle Network hackathon — AI × RWA track.

---

What makes it real (no mock data)

Layer	Source	Used for
Mantle RPC (rpc.mantle.xyz, chain 5000)	on-chain reads via viem	USDY / mETH balances, total supplies, block height
Ethereum RPC	mETH Protocol Staking 0xe3cB…9E8f → mETHToETH()	true mETH→ETH redemption rate (peg monitoring)
DeFiLlama yields.llama.fi	46+ live Mantle pools	APY, TVL, APY volatility (σ), 90-day APY history
CoinGecko	simple/price	live USDY / mETH / MNT / ETH prices
OpenRouter	OpenAI-compatible chat	AI strategy thesis + risk narrative (optional)

Verified token contracts (checked on-chain via symbol()):

• USDY (Ondo U.S. Dollar Yield, Mantle): 0x5bE26527e817998A7206475496fDE1E68957c5A6
• mETH (Mantle Staked Ether, Mantle): 0xcDA86A272531e8640cD7F1a92c01839911B90bb0
• mETH (Ethereum L1): 0xd5F7838F5C461fefF7FE49ea5ebaF7728bB0ADfa

---

Features

1. Live RWA market board — USDY (Ondo T-bill note) and mETH (liquid staking) hero cards with real APY, TVL, on-chain redemption rate, and 90-day APY sparklines, plus a table of all live Mantle yield venues sorted by risk-adjusted APY.
2. AI dynamic yield strategy — pick a risk appetite (Conservative / Balanced / Aggressive); a mean-variance-flavoured optimizer builds a position-capped allocation over real pools, and the AI explains the thesis + emits automated guardrails. Shows projected APY, delta vs current, projected risk score, allocation pie, and concrete rebalance actions.
3. Automated risk management — deterministic, data-driven checks each refresh:
   • USDY NAV/peg — flags trading below accrued floor.
   • mETH ⇄ ETH peg — on-chain redemption rate vs secondary market price.
   • Concentration, idle-capital, and per-pool liquidity/volatility risk scoring.
4. 100% real data, no mocks — connect an injected wallet (MetaMask/Rabby) or paste any Mantle address to watch — both read live balanceOf on-chain. With nothing connected the portfolio panel prompts you, while the strategy + market peg checks still run on real data, so the dashboard is never blank and never fabricated.

Risk model (transparent, deterministic)

Each pool gets a 0–100 risk score from real signals: asset-class baseline, liquidity depth (TVL), yield volatility (σ of daily APY), an excess-yield premium, and incentive dependence. The optimizer maximizes APY / (1 + λ·risk) subject to per-leg caps and a risk budget set by the chosen appetite. The AI layer explains these numbers — it never invents them.

---

Run it

npm install
cp .env.example .env        # optional — works without keys
npm run dev                 # http://localhost:3000

Public RPCs and DeFiLlama/CoinGecko need no API key, so the app runs out of the box in Heuristic mode. To turn on the AI engine, add an OpenRouter key to .env:

OPENROUTER_API_KEY=sk-or-...
OPENROUTER_MODEL=anthropic/claude-sonnet-4.6   # any OpenRouter model slug

The header badge shows AI engine live vs Heuristic mode; every AI output is tagged with its source and gracefully falls back to the deterministic engine on any error.

---

Architecture

src/
  app/
    page.tsx              Dashboard (client orchestrator)
    api/
      markets/route.ts    Live snapshot: prices + on-chain + pools + APY history
      portfolio/route.ts  Live on-chain balances for a wallet/watched address
      strategy/route.ts   Optimizer + AI thesis + rebalances
      risk/route.ts       Risk report + AI narrative
  lib/
    chain.ts              viem clients + on-chain reads (rate, balances, supply)
    datasources.ts        DeFiLlama + CoinGecko fetchers, snapshot assembly
    risk.ts               Asset classifier + deterministic pool risk scoring
    optimizer.ts          Risk-budgeted allocation + rebalance diffing
    riskreport.ts         Peg/concentration/idle automated checks
    ai.ts                 OpenRouter client + heuristic fallbacks
    portfolio.ts          Position valuation (mETH at on-chain redemption rate)
  components/             Dashboard UI (charts, panels, primitives)

Notes / disclaimer

This is a hackathon prototype for demonstration. Allocations and risk scores are informational only and are not financial advice. Always verify contract addresses against official Ondo and mETH Protocol sources before transacting.