All projects
POCAgentic AIBanking / Finance

TrustChain AI

Human-governed, Web3-verifiable autonomous AI agents

Personal project

Runnable proof of concept with mock data, mock payments and a local blockchain.

An AI treasury agent investigates supplier invoices inside a system where it has no authority to move money: policy decides, humans approve, and a blockchain makes the audit record tamper-evident.

Problem & business context

Once an AI agent can act, an organization must be able to answer: who authorized it, what evidence it used, who approved the action, what was executed — and whether the record has been changed since.

My role & contribution

Designed and built the POC.

What I built or designed

  • Treasury agent with read-only tools and RAG over policies and contracts
  • Deterministic governance engine outside the model: identity, permissions, policy and risk → allow, deny or require approval
  • Human approval with roles, limits and four-eyes checks
  • On-chain agent registry and tamper-evident audit records; idempotent mock payment tool