Platform Architecture — The seven modules of Aureon MRM

1

Aureon Inventory — single source of truth.

Every model. Model cards, embedded reports, live performance tagging. Active, inactive, decommissioned — all searchable, all governed.

Every model has a live card. Metadata, methodology, data sources, limitations, assumptions, validation history, performance tags — one canonical record. Version-controlled. Ledger-backed. Ten-year retention.

Search by tier, RAG status, business unit, model type, expiry window. Drill into the full card. Every state change signed and logged. If it's in production, it's in the inventory.

Explore the Model Inventory use case →

Portfolio slice

T1PD Retailv3.2

T1LGD Corporatev2.1

T1IFRS 9 ECLIn review

T2Fraud MLOverdue

+ 123 more · full detail behind demo

2

Aureon Model Lifecycle — onboarding to attestation.

The workflow spine for every model. Every stage AI-copiloted. Every handoff auditable. Nothing falls off.

Lifecycle stages

Model Onboarding

Initial Risk Assessment & Tiering

Independent Validation

4Findings & Remediation

5MRMC Approval

6Deployment & Attestation

7Ongoing Monitoring & Re-attestation

Aureon operationalises the lifecycle the regulators expect — onboarding, tiering, validation, findings, MRMC approval, deployment, attestation, re-validation. Every stage is a first-class step in the platform.

Handoffs auto-route to the right roles — developer, validator, model owner, MRMC. The Aureon Co-pilot drafts artefacts at each stage; humans review and sign. If it slips a deadline, everyone knows.

3

Aureon Validation Studio — dedicated AI workbenches per model type.

Not one generic tool retrofitted for everything. Purpose-built workbenches for Basel PD/LGD/EAD, IFRS 9, scorecards, ML models, marketing propensity — each with the right test suite, benchmarks and evidence template.

Each workbench brings model-specific defaults: mandatory tests, tier-aware benchmarks, methodology templates, calibration thresholds — pre-configured for the model type at hand. The validator focuses on judgement, not setup.

Everything captured to evidence. Everything signed. One workflow, model-aware defaults, zero reconciliation.

Explore the Model Validation use case →

Workbench catalog

Basel — IRB PD/LGD/EAD12 tests

IFRS 9 / Ind AS 109 ECL8 tests

A/B/C Scoring10 tests

Fraud ML9 tests

Marketing & Propensity7 tests

Rating & Corporate8 tests

4

Aureon Challenger Studio — research agents assess and build.

Not a spreadsheet exercise. Research agents assess conceptual soundness, auto-build a challenger model on your data, quantify lift, produce a defensible review — end to end.

Agent pipeline

assess PD Retail conceptual soundness

↓

Research agent reviews methodology, data lineage, comparable models, regulatory guidance…

↓

Challenger built on same data · alternative features + XGBoost baseline

↓

ReadyLift +2.3 Gini pts · defensible review draft

Combines a research agent — reading the model methodology, comparable literature and regulatory guidance to assess conceptual soundness — with a build agent that trains a challenger on the same data and quantifies lift.

Output: a clean, cited review addressing conceptual soundness, alternative specifications and challenger performance. Satisfies RBI MRM 2026 and SR 11-7 effective-challenge requirements.

Explore the Effective Challenge use case →

5

Aureon MRM Coding Agent — deep MRM domain, natural-language input.

An AI capability, not a page. Weaves through every workbench. Understands MRM methodology, statistical patterns, regulatory conventions. Generates code, executes it, returns cited results.

Ask in plain English — "run Hosmer-Lemeshow on the Q3 calibration dataset with 10 bins", or "build a rank-order plot for the PD Retail behaviour segment". The Coding Agent writes the code, executes it against your data, saves plots to the validation package.

Every generation cited to source. Every execution written to the audit trail. Every result reviewable line by line before it counts as evidence.

Prompt → execution

run Hosmer-Lemeshow on Q3 calibration

↓

generate_metric_bundle(
  "hosmer_lemeshow",
  split=Q3_calib,
  bins=10
)

↓

4.2sχ² = 12.4 · p = 0.14 · added to pack

6

Aureon Document Generator — any MRM document, drafted from evidence.

Model docs, validation summaries, executive reports, findings packs, board memos, regulator responses. Drafted from the live inventory and validation runs. Reviewed and signed by humans.

Output types

Model Documentation · SR 11-7 · RBI MRM · APRA CPG 235

Validation Summary · Full evidence pack

Executive / MRMC Report · Board-ready

Findings & Remediation Pack

Regulator Response · Examiner-ready

Every document draws directly from live evidence — inventory, validation runs, findings register, audit trail. No copy-paste from spreadsheets. No stale numbers. The Coding Agent's outputs are the source; the Document Generator narrates.

Track-changes on every revision. Version history tied to a signed author. Final approval is always human — the AI drafts, the human owns.

7

Aureon Executive / MRMC Dashboard — strategic risk at a glance.

Purpose-built for the CRO, CFO and Model Risk Management Committee. Portfolio-wide compliance, tier-weighted exposure, breach concentration — always current. The MRMC pack IS the platform.

Portfolio-wide inventory compliance by tier. Tier-weighted risk exposure. High-severity findings by business unit. Validations pending final approval. Models expiring in 90 days. Regulatory risk flagged.

No more assembling MRMC packs the week before the meeting. No more stale slides. Filters by BU, region, model type — one click to the underlying model, one click to the finding.

Executive view

127Models

Assets under risk$0.1B

Expiring · 90d03

High-severity findings02

Overdue validation03

Seven modules. One AI-native platform.

Seven modules. One workflow.