How SmartPayEngine is built
A non-confidential overview of the architecture: the financial operating layer, the ledger core, the multi-currency data model, the intelligence layer and the security controls. Detailed schemas and source access are shared under a due-diligence process.
Platform layers
Financial data model
Ledger-first balances
Every balance is the sum of its ledger entries. Mutations happen only through controlled transaction operations, which keeps debits and credits paired.
Currency as a first-class field
Accounts, transactions, ledger entries, fees and settlements all carry currency. Conversions post paired cross-currency entries with the quoted rate retained.
Money-movement state machines
Payments, refunds, disputes, chargebacks, settlements and payouts each move through defined states, so an entry can always be explained.
Reversal monitoring
Reversal events are recorded and observable for audit. Reversals are read-only today; refunds and disputes are the implemented recovery workflows.
Intelligence layer
The assistant runs server-side against the Anthropic API with scoped access to the requesting account's financial records. It reads, explains and scores — it does not move money. Any action with financial effect is executed by an operator through the normal controlled operations, with the same authorisation checks as the UI.
- Model access is server-side; no model keys reach the browser.
- Queries are scoped to the caller's account through the same authorisation layer as the API.
- Reversal handling is read-only and observational; the model does not execute reversals.
- Risk scoring, routing intelligence and portfolio analytics are designed but not yet implemented.
Security controls
Implemented
- Authentication with server-side session validation
- Role-based access control from a dedicated roles table
- Row Level Security on every database table
- Privileged operations restricted to the server
- Audit logging of administrative and financial operations
- Hardened HTTP security headers, including CSP and HSTS
Not claimed
- No SOC 2, ISO 27001 or PCI-DSS certification
- No independent penetration test or third-party audit
- No payment-institution licence or registration
- No formal incident-response programme
- MFA available through the identity layer but not enforced platform-wide
Sandbox architecture
All environments run in sandbox/test mode. Processor credentials are test-mode, merchants and cases are seeded, and no transaction represents real value. This makes the platform safe for demonstrations, integration testing, QA, developer experimentation, technical due diligence and pilot preparation — and it means the platform holds no production customer funds or data today.