Housing Help

HUD (U.S. Department of Housing and Urban Development) is the federal agency that funds and regulates affordable housing across the United States. Our primary stakeholder was HUD's Atlanta Regional Office, which oversees federal housing programs across eight southeastern states, including Georgia, Alabama, Florida, and Tennessee.

HUD doesn't operate the housing itself. It subcontracts the day-to-day management and operation of each property to landlords, who are legally responsible for keeping every unit in compliant condition. Inspectors enforce that contract, running annual inspections and responding whenever a deficiency is reported. If a landlord fails to repair within the regulated timeframes, they lose the right to operate the housing.

NSPIRE (National Standards for the Physical Inspection of Real Estate) is HUD's new federal inspection framework. The transition to NSPIRE created a fragmented, high-friction process between inspectors, landlords, and tenants, with no standardized digital platform connecting the three. Deloitte, working with HUD's Atlanta Regional Office, tasked our SCADpro team with designing an AI-powered mobile platform to bridge that gap. Because NSPIRE is a federal standard, the platform is designed to scale to any regulated housing group operating under it, nationwide, not just the Southeast.

As Prototyping Lead and Design Systems Lead, I owned the structural backbone of the product, from component architecture to interactive prototypes. I also contributed to research using AI tools, designed the AI interaction model, and led the resolution of a branding crisis mid-project. My work spanned strategy, systems thinking, and hands-on design execution.

The UX & Research Team

I worked inside a tight UX and research pod embedded in the broader ~25-person SCADpro × Deloitte team.

The inspection ecosystem involves tightly coupled stakeholders with conflicting needs. Our research revealed a vicious cycle: tenants distrust management, so they underreport issues. Landlords receive inaccurate data, so they can't prioritize repairs. Inspectors are buried in documentation overhead. The system erodes itself.

AI wasn't a feature, it was an architectural decision. I defined three core AI functions: a Documentation Assistant that guides photo capture with real-time angle and lighting feedback, a Metric Interpreter that translates regulatory language into plain terms, and a Compliance Decision Support system that predicts violations from historical patterns. The key principle: AI advises, humans decide. Every AI output is subject to inspector confirmation.

We (I and my teammate, Catelynn Kai) started the design system on the first day of the project, intentionally decoupled from final branding. This decision saved weeks of rework later. The system was architected for three user roles, a glass morphism visual language, and national-scale deployment with jurisdiction-level customization.

The app is organized into five modules: Inspection, Remediation, Role-Based Dashboards, AI Assistant Layer, and Analytics. The team designed mid-fidelity wireframes for all three user flows before committing to visual design, validating structure with Google Design experts and Deloitte stakeholders before investing in pixels.

We tested mid-to-high fidelity prototypes with a mix of landlords, tenants, and limited inspector access. Three critical issues surfaced that reshaped the product.

Three role-based experiences unified under one design system. Glass morphism surfaces, AI-guided workflows, and a component architecture built for national deployment.

The role-based color strategy extended beyond the app into a complete brand and marketing system, convention materials, street advertising, office collateral, and promotional products. This demonstrated how a design system decision can scale from UI components to physical brand presence.

Deloitte's organizational structure meant multiple decision-making bodies with competing priorities. Scope creep from housing authorities' wish lists constantly pulled against the need for a focused MVP. I learned to navigate this by presenting design decisions as strategic trade-offs, not aesthetic preferences, which built trust and accelerated consensus.

The complete process book and final presentation deck are available below.

Housing Help shipped as a validated design system and interaction model within a 15-week internship, not a product measured against live tenant or inspector traffic. Where I have a real before/after number, it's labeled measured below. Where the outcome is expert or stakeholder validation rather than a metric, it's labeled qualitative, so the distinction stays honest.

The AI-guidance model went through the same measured iteration. v1 shipped as "user controls, AI validates after the fact." Testing surfaced the confusion directly, "I don't know if the AI is happy with the photo," and v2 shipped as "AI guides in real-time, user confirms." That's a validated design iteration, not a hypothesis, it changed shape because of what a real test participant said.