MUSE & DevJoy — an AI platform for model lifecycle governance
Turning ML training and TechOPS into consumer-grade tooling.
Fannie Mae's model risk management organization operates hundreds of machine-learning models that underwrite trillions of dollars in mortgage decisions. The supporting tooling had sprawled into 14 separate internal applications, each owned by a different team and used by a different persona — model owners, lead model users, auditors and governance specialists — with no shared design language, navigation or data model between them.
Risk officers, data scientists and TechOPS engineers had no shared language. Each handoff lost context, blocked audits and stretched model lifecycle work to multi-week cycles.
- 01
Mapped the end-to-end model lifecycle with 22 stakeholders across risk, data science and operations.
- 02
Defined the MUSE design system with shared tokens, motion and accessibility primitives across 4 product surfaces.
- 03
Designed AI document processing flows that surface model evidence inline rather than buried in artifacts.
- 04
Validated each release with moderated usability sessions and instrumented telemetry against task-success benchmarks.
Mapping a fragmented lifecycle
Workshops with 22 stakeholders surfaced 9 high-friction handoffs. We rebuilt the information architecture around model evidence — the artifact every persona actually cares about.
- 22 stakeholder interviews
- 9 friction points mapped
- 4 personas unified under one IA
MUSE design system
Tokens, motion, accessibility and content patterns ship as a single library consumed by 4 surfaces — from analyst dashboards to TechOPS terminals. Designers and engineers share the same source of truth.
AI document processing
An AI layer ingests model artifacts, summarises evidence and pre-fills governance forms. Risk analysts stay in flow, intervene only on exceptions, and audit trails are generated automatically.
What shipped, and what changed.
Cycle time on model approvals reduced by 50%.
User errors on critical governance flows down 62%.
Recoveries processing collapsed from one week to a few hours.
Single design system adopted across four enterprise products.