Financial AI
you can audit.
Oryze AI builds the intelligence layer inside Oryze — the financial operating platform for entrepreneurs. Every output carries its source. Every projection is labeled. Every decision stays human.
Three rules. No exceptions.
Every number has a source
Each output carries its versioned rule, its computation timestamp and its data freshness — visible to the user, not buried in logs.
It prepares. You decide.
The intelligence prepares reminders, provisions and alerts. It never sends, files or moves anything on its own — the last hand is human, by design.
Fail closed
When the system cannot know, it stops and says so. No plausible answer by default — an honest refusal beats a confident guess.
Not a lab. Shipped.
Four capabilities run today inside the Oryze product demo — a full journey from invoice to accounting export.
Boring where it must be. Sharp where it counts.
Deterministic derivations and versioned rules first. Models only where they add value — never between a number and its source.
Row-level isolation enforced by the database itself. Audit trail and event outbox written in the same transaction as the data.
cash_tension-v1.0, relance-overdue-v1.0 — every rule is named, versioned, and cited by its outputs. Change the rule, the version changes.
A CI of guards and canaries that catch truth regressions — including, more than once, in our own documentation.
Founding engineers, Paris.
We build the intelligence layer of a financial platform — in a culture where every number must prove itself. Small team, real product, hard standards.