Marketing tells people why a product matters. UX makes sure it does. I design with both in mind.
I didn’t take the traditional route into design. I started with , moved into , and became fascinated by how products were sold to me. That curiosity led me to at Drexel, where everything clicked: marketing and UX both start with understanding people.
Today, I bring storytelling, visual thinking, and that understanding into the products I design.
Big family, busy city, and a calling I found by building things.
I grew up in a busy city, in a big extended family. We traveled a lot, so I changed schools often and learned early how to walk into a new room and make friends.
Moving to Canada and getting into project-based work is where I found my calling: building things that make life easier for people.
An AI home maintenance companion that helps homeowners understand a problem before deciding what to do next.
Role
Product Designer & Researcher. Started the project, recruited the team, and led research and design direction.
Team
3 people
Timeline
2 academic quarters
Tools
Figma, FigJam, Figma Make, Google Forms
The problem
Homeowners wanted to understand a problem before hiring anyone.
When something breaks at home, people bounce between Google, YouTube, Reddit and contractor sites, unsure what is wrong or who to trust. Existing platforms skip straight to "find a pro."
5+ platforms to diagnose one issue
No urgency check before spending money
Research
Talking to people and studying the competition.
Research covered interviews and a survey with homeowners, renters and people in shared housing. Anyone who shares a living space deals with home issues: some people fix things themselves and some don't, and renters who keep track of issues are less likely to lose their deposit. Research also included an analysis of competing service platforms and marketplaces, and a map of the DIY resources people turn to.
User interviews
Survey
Competitive analysis
Marketplace review
DIY resource mapping
Every competitor sent people straight to "find a pro" without helping them understand the problem first. The notes were grouped into themes, and two findings set the direction.
Affinity map across four themes.
Finding 01
People couldn't tell how urgent a problem was, or whether they needed a pro at all.
Desired features, grouped from interview notes.
Finding 02
People checked five or more sources to diagnose one issue and still weren't sure how serious it was.
How people research repairs today: scattered sources, unclear severity.
The pivot
The contractor marketplace was the wrong idea.
The brief was to help people find contractors, and the interview questions were built around that process. In answering them, people kept talking about everything that comes before it: working out what's wrong and how serious it is. They needed confidence before deciding whether they needed a pro at all.
BeforeHelp people find and book contractors faster
→
AfterHelp people understand problems and make informed decisions
Designing the AI experience
The user stays in control of the decision.
Someone describes the problem and Sahay explains what's likely going on, then suggests a path that fits their DIY skill level. They decide what happens next. AI can make mistakes, so Sahay never acts for them and the final call stays with the user.
Validation
Flows were validated before screens were designed.
The prototype would be built with AI, so the information architecture and end-to-end flows were tested with participants first, to confirm this was what people wanted. The AI prototyping setup was prepared in parallel, so building could start fast once the flows held up. Participants supported:
Starting with "what's wrong" instead of "find a pro"
Getting a diagnosis before booking anyone
Help deciding, instead of a list of options
Reading the flow
Everything branches from one AI home screen.
Getting in. First-time users go through onboarding: profile, home details, then inviting household members, since homes are often shared by couples or by parents with grown-up kids. Returning users land straight on the AI Assistant home.
Starting with the issue. People open the app to understand an issue quickly, so the AI Assistant is the home screen. A problem can come in as a photo or video, typed text or voice: in an emergency, showing a video is faster than typing, and some problems are easier to talk through.
Clarifying before recommending. Instead of assuming, the AI asks follow-up questions about the problem and the user's DIY skill, then gives an overview. The user chooses: DIY or hire help.
Fixing it yourself. DIY creates a project with how-to steps and a checklist. If the issue is resolved, it's logged in the home's maintenance history. AI can get things wrong, so if the issue isn't resolved, the user can contact a vendor or troubleshoot with the AI again.
Hiring a pro. Hire help matches the user with vendors. Booking runs through ZIP code, contact info and reviews, data-sharing consent, a primary time plus a backup in case the first slot doesn't work out, confirmation and a reminder. The issue then updates with the vendor's details.
The other tabs. The nav bar also opens the issue board (every problem, add new or update existing), DIY guides with short videos to watch or save, the vendor directory, and profile and household settings.
IA and user flows, tested before high-fidelity work.
Final solution
The prototype covers diagnosis, DIY help and booking.
This is the working prototype. Sign up, describe a problem, get a diagnosis, then browse DIY guides and pros.
The first plan collected a lot of home details upfront: the home's age, its systems, past repairs and appliances. In testing, people rushed through it or quit. I redesigned onboarding as a few quick questions that feed MyHome, where AI summarizes the details.
First planA long questionnaire upfront, and people dropped off
→
DecisionA few quick questions, then an AI summary of the home that fills in detail over time
Decision 2: A MyHome hub that's useful on day one
Caring for a home goes beyond fixing issues, so I designed MyHome as a hub instead of a static profile. It shows maintenance reminders, AI insights about the home and quick access to diagnosis. A maintenance record keeps appliances up to date, and it's a selling point when someone sells or rents out the home.
Decision 3: AI writes the home summary
Instead of asking users to document everything, I introduced an AI summary that combines onboarding data, past interactions and home details into one overview. Many participants named it one of the most valuable parts.
Decision 4: Keeping DIY guidance
There was a push to cut DIY guidance. The research showed people want to try simple fixes first, so I argued to keep it and designed it. In later testing, participants often called it one of the strongest features.
Decision 5: Validating flows before building screens
I tested the flows with participants first, and set up the AI prototyping tools in parallel. Once the flows held up, I could build the prototype fast.
Decision 6: Photo, video, text or voice
People describe home problems in different ways, and in an emergency, showing a video is faster than typing. I designed every input to lead to the same diagnosis.
Decision 7: The user makes the final call
AI can make mistakes, so I designed Sahay to ask clarifying questions, suggest a path that fits the user's DIY skill, and never act on their behalf.
The moment it clicked
“This feels like the app actually knows my home.”
A participant in usability testing, on the AI home summary
Outcome
Participants called Sahay "a smart home partner."
They responded well to the AI document summaries, maintenance log, conversational AI and combined DIY and professional help. They were reacting to the vision more than to what the AI could do. People judge an AI product on how well the AI actually works, so what it promises has to match what it delivers.
Next steps
Designing for when the AI is unsure or wrong.
None of this was in the prototype. The open questions:
What happens if the system doesn't have enough information to recommend anything?
How should it say it's unsure instead of sounding confident?
What happens if the user disagrees with the recommendation?
When should it stop asking questions and recommend a professional?
