​
Overview

The MIZU Data Network is designed to enable developers to deploy and run LLM-driven data workflows in a transparent, verifiable, and secure manner. The network is composed of three main components: the core network, the trusted data engine, and the AVS.

​
Components

​
1. The Core Network

The core component of the MIZU Data Network is a decentralized network where all nodes run a DataVM (Data Virtual Machine). The DataVM is similar to the Ethereum Virtual Machine (EVM) but is designed specifically for deploying and running data workflows instead of smart contracts. The workflows could be used for data generation, data validation or data classification.

Developers can deploy and run customized data workflows in MIZU network, which will be compiled to multi-step data tasks. These tasks are then picked up and executed by the edge network. The core network maintains the state of each data workflow and task, ensuring that the processing is carried out in a decentralized and transparent manner.

​
2. The Edge Network

The Edge Network is a critical component of the MIZU Data Network, responsible for executing real data processing tasks. This distributed network comprises a diverse array of devices, including laptops, mobile devices, and dedicated GPU nodes. These devices host large language models (LLMs) or run simulators to perform various data processing tasks efficiently.

​
Verifiable Inference Result

For now we are running our own centralized inference network and simualtor services. All generated results(from inference or simulatora) will be signed by a gateway to ensure that the results are trustable and verifiable. In the future, the network plans to integrate a distributed inference or simulator network to further enhance decentralization and scalability.

​
3. The AVS Layer

The security of the MIZU Data Network is guaranteed by the AVS (Actively Validated Services) Layers. The network follows a standard optimistic strategy for fraud proofs, separating the network nodes into two roles: executors and validators.

Executors are responsible for running data workflows, while validators verify the workflow executions. All MIZU nodes must stake tokens to join the network and can earn yield based on their staking. Users pay gas fees to executors to run their workflows.

The network assumes that all executors are honest, but they are required to generate proofs (merkle proofs or zero-knowledge proofs) for every workflow step and store these proofs on-chain. Anyone can request to challenge any executed step, triggering all validators to re-run the step and vote to slash the staking of malicious executors.

​
Data Storage

The Data Storage Layer is responsible for hosting all intermediary results and outputs of data workflows, such as proofs and final datasets. This layer can be implemented using a Decentralized Autonomous (DA) layer, such as ZeroGravity.

To optimize gas costs, the current design stores proof headers on-chain while storing the proof bodies in the DA layer. The final generated datasets are also hosted in the DA layer, making it easy for users to download, view, or replicate the data.

​
Conclusion

The MIZU Data Network provides a decentralized, secure, and verifiable platform for deploying and running data workflows. By combining the DePIN Network, Verifiable Inference Network, AVS Layers, and Data Storage Layer, the network enables developers to process and analyze data in a transparent and trustless manner, while ensuring the integrity and security of the data workflows through a robust verification and slashing mechanism. The addition of the Data Storage Layer further enhances the network’s capabilities by providing efficient storage and retrieval of proofs and datasets, making them easily accessible to users and other systems.