Skip to content

README.md

Latest commit

 

History

History
61 lines (44 loc) · 2.37 KB

README.md

File metadata and controls

61 lines (44 loc) · 2.37 KB

ModelTrace: Verifiable AI Governance & Auditing

ModelTrace is a specialized protocol built on the Stellar Soroban network, designed to provide transparency, accountability, and auditability for Artificial Intelligence systems. It records a verifiable audit trail of AI model usage, parameter changes, and provenance on an immutable ledger.

The Need for Traceability

As AI models become central to decision-making in finance, healthcare, and law, the "Black Box" problem poses a significant risk. ModelTrace provides a decentralized "Flight Recorder" for AI, ensuring every major decision can be audited and traced back to a specific model version and dataset.

Core Protocol Engines

🔍 Model Registry (contracts/model-registry)

Acts as the authoritative directory for AI models. It stores model architectures, cryptographic hashes of weights, and versioning data.

  • Key Feature: Ensures that a model being called is exactly the version that was audited.

📝 Usage Auditor (contracts/usage-auditor)

Logs the metadata of every inference request and response (anonymized).

  • Key Feature: Provides an immutable timestamp and proof-of-execution for AI services.

🧬 Provenance Engine (contracts/provenance-engine)

Tracks the lineage of a model, including the datasets used for training and the fine-tuning history.

  • Key Feature: Enables ethical auditing of data sources and bias tracking.

Technical Stack

  • Smart Contracts: Soroban / Rust (optimized for high-throughput event logging).
  • Audit Backend: Fastify / TypeScript - Processes and indexes model events for real-time monitoring.
  • Auditor Dashboard: Next.js - A sophisticated interface for AI compliance officers and developers.

Usage Example

// Example of recording an inference event via the ModelTrace API
const modelTrace = new ModelTraceClient(CONFIG);

await modelTrace.recordInference({
  modelHash: "0x123...abc",
  inputHash: "0xdef...456",
  outputHash: "0x789...ghi",
  timestamp: Date.now()
});

Installation & Setup

Contract Build

cargo build --target wasm32-unknown-unknown --release

Developer Dashboard

cd apps/web
pnpm install
pnpm dev

Mission

To foster trust in Artificial Intelligence by providing the decentralized tools required for true governance and ethical accountability.

Bringing transparency to the AI frontier.