Momenta AVS

Demo

About

The problem it solves Voice spoofing and AI-based scam calls are increasingly affecting individuals globally. As AI tools advance, scammers can now mimic the voices of trusted individuals, like family members, using spoofing technology, making it appear that calls are coming from known and trusted contacts. This creates scenarios where victims are manipulated into believing that their loved ones are in distress or facing emergencies. These incidents pose significant threats to people of all ages, leading to emotional and financial harm, and the increasing sophistication of AI makes detection and prevention more challenging. User Interaction and Data Flow ## Momenta AVS Overview Users can interact with the **Momenta AVS** through a custom UI, which allows them to upload audio and video files (.WAV/.MP4). For the purpose of this competition, the UI serves as the primary interface for invoking the **Momenta AVS** service. However, the **AVS** is designed to support multiple interfaces and can be integrated into other applications in the future. ### AVS Architecture The **Momenta AVS** deploys two Docker containers: - **Event Listener Container**: - Interacts with the UI to receive uploaded files. - Performs preprocessing. - Forwards files to the inference container. - **Inference Container**: - Runs AI-powered deepfake voice detection. ### On-Chain Integration Once processed, the results are **signed and committed on-chain** via a **TaskManager smart contract**. The **EigenLayer smart contracts** then handle the reward and slashing mechanisms. The project architecture and development process ## 💻 Code Architecture ### Solidity Contracts - **TangleTaskManager.sol**: The main contract responsible for: - Creating and tracking tasks (`createNewTask`) - Recording responses from operators - Storing inference results (`recordInferenceResult`) - Managing challenges and disputes - **TangleServiceManager.sol**: Handles AVS registration with EigenLayer: - Interfaces with EigenLayer's AVS Directory - Manages rewards distribution - Enforces operator slashing rules ### Rust Components - **src/main.rs**: Entry point with task manager initialization and event watching - **src/lib.rs**: Core implementation with: - Event handling for inference tasks - Task pre-processing - Response submission to blockchain - ECDSA signature handling - **src/context.rs**: Manages Docker containers and AVS context: - Initializes and configures Docker containers - Handles dynamic port mapping - Manages container lifecycle - Provides shared context for the AVS ## 📖 Smart Contract Events The system emits several key events: - **NewTaskCreated**: When a new audio file is submitted for analysis - **InferenceResultRecorded**: When an inference result is recorded with its prediction and confidence ## 🔄 Docker Container Integration The AVS manages Docker containers automatically: ### Container Configuration: - Containers are pulled from the Docker Hub if not available locally - Each container is assigned to the "eigenavs" Docker network - Ports are dynamically assigned to avoid conflicts ### Container Communication: - The AVS communicates with the checking container via HTTP - The checking container forwards requests to the inference container - Results are returned to the AVS via HTTP responses ### Failure Handling: - Container failures are detected and handled - Automatic cleanup on AVS shutdown prevents resource leaks Product Integrations The project is built and deployed with Tangle Blueprints for modularization. Key differentiators and uniqueness of the project ## 🔑 Key Features & Unique Aspects of Momenta AVS Momenta AVS is a **first-of-its-kind decentralized solution** for audio deepfake detection, leveraging **EigenLayer** and **Tangle Network** to ensure verifiability, security, and transparency. Unlike any other web2.0 AI-based deepfake detection systems, Momenta AVS provides a **trustless, on-chain verification mechanism** that prevents tampering and censorship. ### 🚀 **Key Features** - **Decentralized & Trustless Verification** - Audio files are processed through multiple independent operators, ensuring no single entity controls the verification results. - **On-Chain Reputation & Immutable Storage** - Inference results, including confidence scores, are stored immutably on-chain, allowing platforms, legal entities, and users to reference a **verifiable source of truth**. - **Restaking Mechanism for Security & Economic Incentives** - EigenLayer's **restaking mechanism** ensures economic alignment, incentivizing accurate reporting while penalizing dishonest behavior. - **Tamper-Proof & Transparent Processing** - The use of **ECDSA signatures** and cryptographic proofs ensures that deepfake detection results cannot be manipulated or censored. - **Scalable, Open, & Modular Architecture** - Built with a **modular Docker-based infrastructure**, allowing seamless integration with various applications and enabling scalability across different AI models. Momenta AVS pioneers a **decentralized future for audio deepfake detection**, making AI verification more **trustworthy, transparent, and economically sustainable** for a wide range of applications. Trade-offs and shortcuts while building A basic UI is deploy for interaction with the service, but this will be adjusted to handle other interfaces like: AI Agents, social media interactinos. The signatures used are EDCSA, but will be converted to BLS for scaled aggregation.