muthuk1/mantle-rwa-yield-router

🏗️ Dynamic RWA Yield Router

Mantle Turing Test Hackathon 2026 — Phase 2 "AI Awakening" — AI × RWA Track

An autonomous AI agent that dynamically allocates capital across Mantle's Real World Asset (RWA) stack using reinforcement learning + LSTM yield prediction, with MEV-protected execution, auto-compounding, social sentiment analysis, and all decisions recorded on-chain via ERC-8004 agent identity NFTs. Ships with a Binance-grade Next.js dashboard.

---

🔥 What Makes This 10x Different

Innovation	Description	Status
Ensemble AI Engine	PPO RL optimizer + LSTM yield predictor + sentiment analyzer working in concert	✅ Active
MEV Protection Layer	Sandwich detection, trade splitting, private mempool routing, optimal slippage	✅ Active
Auto-Compounding	Gas-optimized yield restaking with mathematically optimal compound frequency	✅ Active
LSTM Yield Predictor	7-day yield forecasting with MC Dropout confidence intervals & regime detection	✅ Active
Social Sentiment	Real-time crypto sentiment aggregation (Fear & Greed, social volume, news)	✅ Active
ERC-8004 Agent Identity	Soulbound NFT with capabilities, attestations, reputation scoring, decision history	✅ On-chain
zkML-Ready Architecture	Verifiable inference proofs for on-chain AI transparency (EZKL/Giza compatible)	🔜 Planned
Binance-Grade Dashboard	Next.js + Tailwind + Recharts with full Binance design system (dark theme, trading UI)	✅ Built

---

📊 Portfolio Assets

Asset	Type	Target APY	Risk	Protocol
USDY	Tokenized US T-Bills	~4.25% + lending	Low	Ondo Finance
mETH	Liquid Staked ETH	~3.9% + lending	Medium	Mantle LSP
MI4	Tokenized Index Fund	~5.4% + lending	Medium	Securitize

The agent captures base yield + lending yield (Aave V3 supply) for each asset, achieving total APYs of 4.7-6.2%.

---

🏛️ Architecture

┌──────────────────────────────────────────────────────────────────────────┐
│                     DYNAMIC RWA YIELD ROUTER v2.0                        │
│                                                                          │
│  ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌────────────┐       │
│  │  OBSERVE   │→│   PREDICT  │→│   REASON   │→│    PLAN    │       │
│  │            │  │            │  │            │  │            │       │
│  │ DeFiLlama │  │ LSTM Yield │  │ PPO Policy │  │ Rebalance  │       │
│  │ CoinGecko │  │ Predictor  │  │  + Ensemble│  │ + MEV Opt  │       │
│  │ Mantle RPC│  │ + Sentiment│  │            │  │ + Splitter │       │
│  │ FRED/News │  │ + Regime   │  │            │  │            │       │
│  └────────────┘  └────────────┘  └────────────┘  └────────────┘       │
│       ↓                                                ↓               │
│  ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌────────────┐     │
│  │  AUTHORIZE │→│  EXECUTE   │→│AUTOCOMPOUND│→│   VERIFY   │     │
│  │            │  │            │  │            │  │            │     │
│  │ 6-Layer    │  │ Unsigned Tx│  │ Gas-Optimal│  │ On-chain   │     │
│  │ Risk Mgr   │  │ + MEV Prot │  │ Restaking  │  │ + ERC-8004 │     │
│  │ + Sentiment│  │ + Whitelist│  │ Automation │  │ Attestation│     │
│  └────────────┘  └────────────┘  └────────────┘  └────────────┘     │
│                                                                        │
│  ┌──────────────────────────────────────────────────────────────────┐  │
│  │                    ON-CHAIN CONTRACTS                             │  │
│  │  YieldRouterAgent.sol │ AgentIdentity8004.sol │ RiskRegistry.sol│  │
│  └──────────────────────────────────────────────────────────────────┘  │
│                                                                        │
│  ┌──────────────────────────────────────────────────────────────────┐  │
│  │              BINANCE-STYLE NEXT.JS DASHBOARD                     │  │
│  │  Portfolio │ Markets │ AI Optimizer │ Risk │ Agent Identity      │  │
│  └──────────────────────────────────────────────────────────────────┘  │
└──────────────────────────────────────────────────────────────────────────┘

8-Stage Agent Pipeline (v2.0)

1. OBSERVE — Fetch real-time yield, price, macro, and sentiment data
2. PREDICT — LSTM yield forecasting with MC Dropout confidence + regime detection
3. REASON — PPO policy + ensemble (combine RL action with LSTM prediction + sentiment)
4. PLAN — Compute trades with MEV analysis (sandwich detection, trade splitting)
5. AUTHORIZE — 6-layer risk manager (depeg, volatility, concentration, drawdown, circuit breaker, sentiment)
6. EXECUTE — Construct MEV-protected unsigned transactions (private mempool, optimal deadline)
7. AUTO-COMPOUND — Gas-optimized yield restaking at mathematically optimal frequency
8. VERIFY — Confirm on-chain, update ERC-8004 reputation, generate strategy report

---

🧠 AI/ML Stack

PPO Reinforcement Learning

• Algorithm: Proximal Policy Optimization with Actor-Critic MLP
• State Space: 18-dimensional (yields, prices, macro, risk signals, portfolio weights)
• Action Space: 3-dimensional continuous → softmax → weights [USDY, mETH, MI4]
• Reward: Risk-adjusted Sharpe ratio - drawdown penalty - depeg penalty - gas costs

LSTM Yield Predictor (Novel)

• Architecture: 2-layer LSTM with attention mechanism
• Input: Yield history, price momentum, fed rate, volatility, sentiment
• MC Dropout: 50 forward passes for uncertainty estimation (confidence intervals)
• Regime Detection: Bull/Bear/Sideways classification head
• Feature Importance: Attention weights reveal which features drove the prediction

Social Sentiment Analyzer (Novel)

• Sources: Fear & Greed Index, social volume, news sentiment
• Integration: Sentiment score feeds into both LSTM predictor and risk manager
• Impact: High fear → increase USDY (safe haven), High greed → lean into mETH/MI4

---

🛡️ Novel Risk & Execution Features

MEV Protection Layer (Novel)

• Sandwich Attack Detection: Estimates price impact per trade
• Trade Splitting: Automatically splits large trades to reduce MEV exposure
• Private Mempool: Routes high-impact trades through private channels
• Optimal Deadlines: Shorter deadlines for high-risk trades, longer for low-risk

Auto-Compounding Engine (Novel)

• Optimal Frequency: Calculates gas-optimal compound interval using calculus optimization
• Formula: n* = √(APY × Principal / (2 × GasCost)) compounds per year
• Break-even: Computes minimum principal where compounding beats simple yield
• Smart Triggers: Only compounds when accumulated yield > 3× gas cost

6-Layer Risk Management

Layer	Description	Trigger
Depeg Detection	USDY/USD and mETH/ETH peg monitoring	>0.5% / >2%
Volatility Guard	Dynamic exposure reduction in high-vol regimes	ETH 30d vol >80%
Concentration Limits	Position size enforcement	>60% or <5%
Smart Contract Risk	Protocol-weighted risk scoring	Per-protocol scores
Drawdown Protection	Circuit breaker on portfolio losses	DD >10%
Sentiment Guard (Novel)	Reduces risky exposure on extreme fear	Score <20

---

🎨 Dashboard (Next.js + Binance Design)

Full-featured financial dashboard built with the Binance Design System:

• Dark canvas (#0b0e11) with card surfaces (#1e2329)
• Binance Yellow (#FCD535) accent for CTAs and brand elements
• Trading green/red (#0ecb81/#f6465d) for price direction semantics
• BinancePlex substitute (JetBrains Mono) for all financial numbers
• 6 dashboard sections: Portfolio, Markets, AI Optimizer, Risk Monitor, Reports, Agent Identity

Dashboard Features

• 📈 Real-time portfolio value chart with timeframe selector
• 🥧 Live allocation pie chart + weight history area chart
• 📊 RWA markets table with yield comparison bar chart
• 🧠 LSTM yield predictions with confidence bars + AI recommendations
• 🌡️ Social sentiment gauge with source breakdown
• 🛡️ Risk score ring, protection layer status, activity log
• 🤖 ERC-8004 agent identity card with reputation, capabilities, attestations
• 📱 Fully responsive (mobile hamburger, tablet 2-up, desktop full layout)

# Run the dashboard
cd dashboard && npm install && npm run dev

---

🚀 Quick Start

# Clone from HuggingFace
git clone https://huggingface.co/muthuk1/mantle-rwa-yield-router
cd mantle-rwa-yield-router

# Install Python dependencies
pip install -r requirements.txt

# Run demo (3 cycles with live data)
python scripts/demo.py

# Train RL agent
python scripts/train.py

# Run continuous agent
python -m agent.main --mode run --interval 3600 --capital 100000

# Run tests (25 passing)
pytest tests/ -v

# Run dashboard
cd dashboard && npm install && npm run dev

---

📂 Project Structure

mantle-rwa-yield-router/
├── agent/
│   ├── main.py              # 8-stage orchestrator pipeline
│   ├── data_pipeline.py     # Real-time data aggregation (8 sources)
│   ├── rl_optimizer.py      # PPO agent + Gymnasium environment
│   ├── risk_manager.py      # 6-layer risk management
│   ├── executor.py          # MEV-protected unsigned tx builder
│   ├── strategy_reporter.py # LLM strategy letter generator
│   └── novel_features.py    # LSTM predictor, MEV shield, auto-compound, sentiment
├── contracts/
│   ├── YieldRouterAgent.sol  # On-chain allocation router
│   ├── AgentIdentity8004.sol # ERC-8004 soulbound agent identity
│   └── RiskRegistry.sol      # On-chain risk parameters
├── dashboard/                # Next.js Binance-style UI
│   ├── src/app/page.tsx      # Main dashboard page
│   ├── src/components/       # TopNav, Charts, RiskMonitor, AIOptimizer, AgentIdentity
│   └── src/lib/data.ts       # Data generators and mock data
├── telegram_bot/
│   └── bot.py               # Telegram UI (/status, /yields, /risk)
├── config/
│   └── constants.py         # All contract addresses, ABIs, parameters
├── scripts/
│   ├── demo.py              # Demo runner
│   └── train.py             # RL training script
├── tests/
│   └── test_agent.py        # 25 unit + integration tests
├── requirements.txt
├── .env.example
└── README.md

---

📜 Smart Contracts

YieldRouterAgent.sol

On-chain allocation records, position limits, circuit breaker, operator authorization.

AgentIdentity8004.sol (ERC-8004)

Soulbound agent identity NFT: capabilities, attestations, reputation (0-10000), decision history.

RiskRegistry.sol

On-chain risk parameters, depeg event tracking, allocation validation.

---

📊 Contract Addresses (Mantle Mainnet, Chain ID 5000)

Contract	Address
USDY (Ondo)	0x5bE26527e817998A7206475496fDE1E68957c5A6
mETH	0xcDA86A272531e8640cD7F1a92c01839911B90bb0
USDC	0x09Bc4E0D864854c6aFB6eB9A9cdF58aC190D0dF9
WMNT	0x78c1b0C915c4FAA5FffA6CAbf0219DA63d7f4cb8
Fluxion Router	0x5628a59df0ecac3f3171f877a94beb26ba6dfaa0
Agni Router	0x319B69888b0d11cEC22caA5034e25FfFBDc88421
Aave V3 Pool	0x458F293454fE0d67EC0655f3672301301DD51422

---

🔑 Key Design Decisions

1. Construct-Only Safety — Never holds private keys. All outputs are unsigned tx payloads.
2. Ensemble Intelligence — PPO + LSTM + Sentiment provide richer signals than any single model.
3. MEV-Aware — Every trade goes through sandwich detection + optimal splitting.
4. Gas-Optimized Compounding — Mathematical optimization of compound frequency vs gas costs.
5. On-Chain Audit Trail — ERC-8004 records every decision with content hashes.
6. Binance-Grade UI — Production dashboard, not a demo. Full dark theme, live charts, responsive.

---

🏆 Hackathon Track

Mantle Turing Test Hackathon 2026 — Phase 2 "AI Awakening"

• Track: AI × RWA ($100K prize pool)
• Focus: Autonomous AI agents managing tokenized real-world assets
• Key Tech: RL optimization, LSTM prediction, ERC-8004, MEV protection, Mantle L2

📝 License

MIT

---

Built with 🤖 for the Mantle Turing Test Hackathon 2026