A Verifiable AI Platform

1. Mission & Vision

Mission Statement

To build a decentralized platform for an AI Ownership Economy. The system will enable individuals and organizations to contribute diverse data types, receive a verifiable share of the IP in the AI models created, and benefit from the resulting scientific, civic, and commercial applications.

Vision

The platform will be a multi-modal data cooperative, launching with three distinct data streams to prove its versatility: Biometric Data (fNIRS), Civic Action Data (ZK-Email), and Behavioral Data (Image Metadata & ZK Mopro).

Core Objectives

• Develop a Multi-Modal B2B Data & AI Platform: Create tools that solve data scarcity across multiple high-value domains.
• Pioneer Verifiable & Private AI: Engineer a trustless, on-chain chain of evidence from raw data to model output, with a clear path towards privacy-preserving computation using FHE.
• Establish a Fair Economic Model: Implement a verifiable system for algorithmically scoring data contributions and distributing IP/revenue.

2. Project Scope: Multi-Modal Data Streams

The platform is being developed with three initial data streams to demonstrate its capabilities:

• Data Stream 1 (Biometric - fNIRS): Researching fNIRS data collection methodologies and building models. The code for this stream is the primary content of the repository in its current state.
• Data Stream 2 (Civic Action - ZK-Email): Developing a system for verifiably rewarding pro-healthcare advocacy using on-device AI and ZK proofs.
• Data Stream 3 (Behavioral - Image Metadata): Designing a feature to incentivize positive sleep habits using on-device AI and ZK Mopro.

3. Technical Architecture

The architecture is designed for verifiability, privacy, and fair value attribution.

Core Technology Stack:

• Frontend/Mobile: Next.js
• Decentralized Storage: Swarm
• Blockchain & Verifiable Compute: NEAR Protocol and Shade Agents
• ZK Proving Systems: ZK-Email, EZKL, ZK Mopro
• Privacy-Preserving Tech (Future): Fully Homomorphic Encryption (FHE)

Verifiable AI & Contribution Scoring Engine:

The core of the platform is a system that provides a chain of evidence for both model accuracy and fair economic distribution. This is achieved using Shade Agents on NEAR, which run containerized Python scripts (for model training and scoring) in a Trusted Execution Environment (TEE). Contribution value is calculated using a Shapley value algorithm, ensuring a trustless and fair record of data value.

4. Project Status & Repository Purpose

This repository is being repurposed to serve as the development hub for the Verifiable AI Platform described above. Initially, it housed a secure file-sharing application (see "Original Project Context" below). The focus has now shifted entirely to the new platform's goals.

Currently, the repository contains the foundational code for Data Stream 1 (Biometric - fNIRS). The primary work is located in the data_and_python directory, which includes scripts for data analysis and machine learning.

For instructions on how to run the fNIRS data analysis pipeline, please refer to the detailed guide in the data_and_python/README.md file.

5. License

This project is licensed under the MIT License.

Original Project Context: Secure File Sharing Application

FileShare - Secure File Sharing Application

A secure file sharing application built with Next.js, React, and TailwindCSS that integrates with Codex and Waku protocols.

Features

• Secure file sharing with end-to-end encryption
• Integration with Codex and Waku protocols
• Real-time status monitoring of Codex and Waku nodes
• Customizable API endpoints
• Modern, responsive UI with dark mode
• Optional client-side encryption powered by TACo with customizable on-chain access conditions
• Seamless Ethereum wallet connectivity (MetaMask) for signing encryption/decryption operations

Codex Integration

The application integrates with Codex nodes through a dedicated client implementation. The integration includes:

• Real-time status monitoring of Codex nodes
• Customizable API endpoint configuration
• Visual indicators for node status (active/inactive)
• Error handling and reporting

Codex Client Architecture

The Codex integration is implemented using a modular architecture:

1. CodexClient Class and useCodex Hook (hooks/useCodex.ts):

   • Contains the complete Codex implementation in a single file
   • Handles all API requests to Codex nodes
   • Manages node status checking and caching
   • Provides methods for interacting with Codex APIs
   • Implements React hook for using the Codex client in components
   • Manages state for node status, loading, and errors
   • Provides methods for updating the API endpoint

2. UI Integration:

   • Visual indicators for node status
   • Form for configuring the API endpoint
   • Error reporting and status messages

TACo Encryption & Access Control

The application uses TACo for client-side encryption and on-chain access control. The integration is encapsulated in hooks/useTaco.ts and is tightly coupled with the file-upload flow.

Key points:

• Files/plaintexts are encrypted in the browser before they are uploaded to Codex or transmitted via Waku – i.e. only ciphertexts ever leave the client.
• Decryption happens on-demand when the data consumer/recipient pings a cohort of TACo nodes. Decryption material is only provisioned when the data consumer (in this example, authenticated via the signature of a connected wallet) satisfies the pre-specified TACo Conditions.
• Helper utilities are provided for two starter TACo conditions types:
  • Positive balance of POL (Polygon Amoy).
  • Time-limited access (N seconds from upload).
• You can define any arbitrary on-chain condition by instantiating a TACo Condition in your code – see TACo docs for more condition logic and types.
• Wallet connectivity (handled transparently by WalletProvider and the Connect Wallet button) is only required to sign TACo encryption/decryption requests and authenticate the data producer and consumer respectively.
• You can define various authentication methods and combine them with conditions, including Sign In With Ethereum and even off-chain identities.

Getting Started

1. Clone the repository
2. Install dependencies: npm install
3. Run the development server: npm run dev
4. Open http://localhost:3000 in your browser

Configuration

Codex Node

By default, the application connects to a Codex node at http://localhost:8080/api/codex. You can change this in the settings panel.

Waku Node

By default, the application connects to a Waku node at http://127.0.0.1:8645. You can change this in the settings panel.

Ethereum Network

The application targets the Polygon Amoy testnet (80002). If your MetaMask wallet is on a different network, the app will automatically request to add/switch to the correct network.

License

MIT