MNEE Sentinel

Project Story

🛡️ Inspiration

The B2B payments market is plagued by a "trust deadlock." Freelancers fear non-payment, and businesses fear receiving subpar work or being scammed. We built MNEE Sentinel to bridge this gap by creating an autonomous AI treasurer that uses math (ZK-proofs) and smart mandates (EIP-712) to ensure fairness for both parties.

🏗️ What it does

MNEE Sentinel is a cross-platform Treasury Agent (Telegram & Slack) that automates B2B procurement:

• Autonomous Negotiation: Gemini AI negotiates discounts with vendors directly in the chat.

• ZK-Verified Escrow: Payments are locked in MNEE and only released once the AI verifies the work matches the mandate.

• Gasless UX: Users never need to hold ETH; all transactions are sponsored by our custom relayer pattern.

• Adversarial Governance: A "Red Team" AI auditor (GeminiAuditor) checks every transaction for policy violations before execution.

🧱 How we built it

Technology Stack

🏛️ Technical Architecture

+-----------------------+
|   MNEE SENTINEL BOT   |
| +----------+ +------+ |
| | Telegram | |Slack | |
| +----+-----+ +--+---+ |
|      |          |     |
|      v          v     |
| +-------------------+ |
| |   SENTINEL CORE   | |
| | (Intent & Logic)  | |
| +---------+---------+ |
|           |           |
|     +-----+-----+     |
|     v           v     |
| +-------+   +-------+ |
| |Azure  |   |Gemini | |
| |Voice  |   |Auditor| |
| +---+---+   +---+---+ |
|     |           |     |
|     +-----+-----+     |
|           |           |
|           v           |
| +-------------------+ |
| |  RELAYER ENGINE   | |
| | (Gasless/EIP-712) | |
| +---------+---------+ |
|           |           |
|           v           |
| +-------------------+ |
| |  SEPOLIA CHAIN    | |
| | (MNEE Settlement) | |
| +-------------------+ |
+-----------------------+

🚧 Challenges we ran into

• The "Paranoid" Auditor: Our Red Team AI was so effective it initially blocked high-value demo payments as "potential laundering." We built Smart Escalation Logic so that "limit hits" automatically trigger the Multi-Sig Flow, turning a hard block into a governance opportunity.

• Docker "Zombie" Processes: Managing dual-bots in one container led to orphan processes. We solved this with concurrently --kill-others to ensure the container self-heals if one bot fails.

• Audio Codec Mismatch: Slack (.webm) and Telegram (.ogg) send different formats. We implemented dynamic Content-Type detection to ensure Azure Speech reached ~90% accuracy.

🏆 Accomplishments that we're proud of

• Security-First Architecture: As a Cybersecurity Graduate, this project allowed me apply the principle of Defense-in-Depth (Biometrics → AI Auditor → Multi-Sig) rather than just delivering a functional prototype.

• Real On-Chain Transactions: Live MNEE transfers on Sepolia with verifiable Etherscan receipts; Generated in the app without having to leave.

• Multi-Sig Emoji Consensus: High-value transactions require a quorum of "✅" reactions from authorized admins before execution.

• Adversarial AI Governance: Our internal Red Team AI actively monitors and prevents fraudulent intent in real-time.

• Voice-to-Intent: Processing biometric voice commands into cryptographically signed mandates via Azure.

• Enterprise Audit Trail: JSON-LD structured logs that are fully exportable to CSV for corporate compliance.

🚀 What's next for MNEE Sentinel

• Mainnet Deployment: Moving from Sepolia to Ethereum Mainnet for live MNEE transactions.

• Advanced ZK-Proofs: Implementing full Groth16 proof generation for private invoice verification.

• Advanced Multi-Sig: Adding time-locked approvals and threshold-weighted voting for large treasury funds.

Built With

• azure-container-apps
• azure-speech-services
• docker
• eip-712
• ethers.js
• gemini-ai
• github-actions
• grammy
• groth16
• node.js
• playwright
• postgresql
• slack-bolt
• sqlite
• typescript