Trust / How it works
How it works
Three platforms, one operating model: data moves one way, the AI reads finished figures, and every change reaches production through a reviewed, tested path.
The three platforms.
The accounting analytics platform connects to a client's accounting and bank data, reconciles it through software we write, and presents it as a secure dashboard plus an AI analyst.
WealthVault consolidates the per-product statements an advisory firm's advisers upload into one client view. Extraction is deterministic parsing; no language model reads identifiable client data. AI analysis runs on the firm's own AI workspace over a tokenised view, through a single read-only connector. The platform holds fabricated data until its go-live gates are met: the responsible party's sign-offs, the product provider's approval, and an independent penetration test.
The trading tool tracks an investment manager's fund trades and flows, reconciled to the fund administrator's own net asset values. It runs as a single-tenant deployment in its own cloud subscription, so it can be handed to the client's IT as one unit.
The data moves one way. The AI reads finished figures and never writes them.
Figures flow from the source system, through reconciliation or extraction, to the dashboard. The AI reads through a read-only database role and cannot author or change a figure; a write attempted through the connector is refused and logged. The AI makes no automated decisions.
Every figure traces back to its source.
On the accounting platform the numbers are inspectable SQL against your own reconciled data, and each figure traces to the source record it came from. On WealthVault every extracted record is tied to the document it was read from.
Access is managed, denies by default, and stores no password of yours.
Sign-in is against a closed allowlist, with no self-registration and no password store of ours. The accounting platform (Auth0) and WealthVault (AWS Cognito) use a one-time email code. The trading tool uses Microsoft Entra ID: people sign in with the Microsoft work accounts their own IT issues, so their organisation's multi-factor and conditional access policies apply, and joiners and leavers stay in their hands. Route-level middleware answers an unauthenticated caller only on a route explicitly marked public. AI connections authenticate with per-deployment tokens, and a startup self-test checks the auth configuration on every deploy.
Every change reaches production through a reviewed, tested path.
All code is in version control. Changes go through pull requests with required CI checks on protected main branches; secret scanning runs over full repository history, and dependency alerts are triaged under a written patch SLA. On WealthVault, deployment gates probe the live environment on every build and fail the build if the posture drifts. Deployments are deliberate, not automatic. Production workloads take their secrets from a managed secrets manager by runtime injection.