Compliance and risk AI built for regulated banks

  • A controlled AI environment, not a free-running agent
  • Runs where your data already lives, on-premises or in your own cloud
  • Your staff make every decision, and every decision is recorded
YOUR ENVIRONMENTYour dataNUCLEUSYour rules checkNamed employeeYour LLM + agentsAUDIT TRAILNothingleaves

Controlled by design, not a free-running agent

A free-running agent decides what to do next and acts on it. NUCLEUS agents propose; they never act: your rules do the checking, LLMs and agents handle what rules cannot, and your staff approve every result.

  • Who decidesDeterministic rules do the checking. LLMs and agents interpret documents, gather data and propose changes. A named employee approves every result.
  • RepeatabilitySame input, same result, every time. LLM and agent steps are scoped to one task and shown for review.
  • ExplainabilityThe rule that fired and the data it used, shown with every result.
  • Data accessDefined tools and data, inside your environment, through an LLM your institution provides.
  • Audit trailInputs, every tool an agent called, rationale, decision and the deciding employee, recorded for every case.
  • Cost10–15× lower token costs on comparable CLM (KYC) tasks.
PLAIN AGENTIC SETUPStaffNUCLEUSRulesLLMStaff
  • Plain agentic setup: picks its own steps, takes longer, and can end differently on the next run
  • NUCLEUS: a fixed path through your rules and scoped AI steps; same input, same result

On-premises or in your own cloud. Nothing transferred.

NUCLEUS is installed inside your environment and works with your own LLM, data stores and workflow tools.

  • Dedicated deployment inside your environment, on-premises or in your own cloud: containerised software for Kubernetes or OpenShift, deployed with Helm
  • No shared multi-tenant environment
  • LLM calls go to an LLM your institution provides, in your own cloud or self-hosted
  • No training on your data
  • PQ Technologies keeps no copy of your data
YOUR BANKFrontOfficeBusiness RiskManagementServiceHubsComplianceNUCLEUSIntelligence Layer on top of what you already runAPI· seamless, real-time integrationCorebankingCRMDocsScreeningLLMvendorAny othersourceYOUR EXISTING SYSTEMS & DATA · NO MIGRATIONClient data never leaves the bank

Tested, encrypted and logged

Vulnerability management

  • Regular external security assessments
  • Automated dependency and container image scanning on every change and release

Identity & encryption

  • Single sign-on via OpenID Connect against your identity provider
  • Roles synced from your IdP, with object-level permissions
  • TLS throughout; APIs secured with OAuth2

Data control

  • Structured, request-bound audit logging
  • Data purged after the retention period you define
  • Your own private image registries supported

Built within the boundaries regulators expect

Two kinds of rules apply to compliance AI in a bank: rules on data protection and resilience, and the rules on AI itself. NUCLEUS is designed around both.

Data protection & resilience

Supported by how NUCLEUS is deployed and who decides

  • DORAEU digital operational resilience For EU institutions: a dedicated deployment inside your environment, shared with no one, and no client data held by PQ Technologies.
  • GDPR & FADPGDPR Art. 22 · FADP Art. 21 No solely automated individual decisions: NUCLEUS proposes, a named employee decides.

AI governance

Supervisory guidance and the EU AI Act, addressed by design

  • FINMAGuidance 08/2024: Governance and risk management when using artificial intelligence Supports FINMA's expectations on AI governance: processing stays with your data, every result can be explained, and a named employee decides.
  • EU AI ActRegulation (EU) 2024/1689 For EU institutions: human oversight, transparency and a record of every AI step, built into how NUCLEUS works.
  • FCAGood practice on AI in financial services (2026) Supports the FCA's good practice on AI: clear accountability, with a named employee for every decision, and staff who can see and challenge every AI result.
  • MASGuidelines on AI Risk Management (2026) Supports MAS's AI lifecycle controls: transparent results, human oversight of every decision, and a record of every AI step for audit.
  • FSBSound Practices for Responsible Adoption of AI (2026) Supports the FSB's sound practices on explainability and human oversight: every result can be explained, a named employee decides, and AI runs inside your environment.

Compliance practice and engineering depth

Since 2001

Compliance expertise through PQ Solutions

PQ Solutions is our expertise partner. Since 2001, it has advised 400+ banks and financial institutions in Switzerland, Liechtenstein, the EU, the UK and the US. That practice is encoded in every Intelligence Package, and its team tracks regulatory change to keep it current.

PQ Solutions
ETH Zurich

Technical expertise from ETH Zurich

PQ Technologies is an ETH Zurich start-up. Our AI engineers build the Intelligence Layer: explainable AI, enterprise integration and governance. ETH Zurich studies confirm 50%+ less manual review effort with NUCLEUS, proven in production at banks.

Our team

See NUCLEUS on your own cases

Walk through your compliance and risk processes with our team.