The intelligence layer above your stack.
Nexara's architecture is formalised in a published, open-access research paper. It is citable, it is open to challenge, and it was written before the platform was built — not to justify it afterwards.
The paper describes a compound AI system for regulated counterparty credit risk. Its central claim is empirical rather than promotional: supervisory requirements, accumulated over fifteen years and written by people with no interest in machine learning, have been implicitly specifying a particular cognitive architecture.
Read together — model validation, risk data aggregation, human oversight, bespoke stress testing — these requirements do not describe a monolithic model. They describe a system with separable, independently auditable components, an explicit memory of past events, and a governed boundary between what it decides alone and what it refers to a person.
That is the architecture Nexara implements. The paper is the specification; the platform is the proof. Publishing it first was a deliberate choice: a claim that can be cited can also be refuted, and a vendor unwilling to be refuted is asking for trust it has not earned.
“Regulated counterparty credit risk has, under very different constraints, converged on the architectural principles the research community is now articulating. A novel empirical observation about the relationship between supervisory requirements and cognitive architecture.”From World Models Under Regulatory Constraint · Weiller, 2026
Citable, open, permanent.
Published open access under CC BY 4.0 with a permanent DOI. Anyone can read the reasoning, test the claim, and disagree in public — including a prospective client's own model validation team.
Discuss the paper directly.
A briefing includes a review of the research with its author, alongside the architecture walkthrough and a live product session.