4th year Engineering Capstone Project - A cumulative task to demonstrate all systems design best practices and apply knowledge gained through the program in a final 8 month group capstone project.

A product motivated by the complex home buying process, untrustworthy real estate agents and the millennial dream of owning property someday.

Disclaimer: All thoughts and opinions expressed in this case study are in collaboration with Team 08 in the Systems Design Engineering Cohort of 2021

My Role

Problem Definition

We get it, buying a home is hard! Homes are rising in price and It takes a lot of money to buy a house so you have to make sure you're doing it right the first time. But for most people who are buying their first homes, they don't have the expertise to navigate through the complex process or their ideal preferences.

Users end up feeling pressured, frustrated and overwhelmed during the home buying process. Resulting in new home owners temporarily giving up and discouraged from the idea of home ownership.

Design Challenges

Data Paywalls

There is no open source database of frequently detailed home listings. Unfortunately a majority of MLS data is barred unless a registered real estate agent has access to the database.

An intuitive Onboarding

We need to establish a balance between collecting personal and financial information to make initial recommendations (the cold start problem)

Recommendation Engine

How can we incorporate a recommendation engine that is not only accurate but trustworthy that user’s will contribute their data towards?

Product Goals

Holistic Experience

Collection of both financial and personal preferences to take into account when making recommendations. Realistic housing options that are within budget and in reason of buyer preferences.

Unbiased Listings

Listings surfaced to the user contain a reduced bias free from external influence and sponsors. Listings are best fit to a user’s profile depending on the information provided

Evolving Recommendations

Recommendations change depending on a user’s changing preferences from implicit and explicit data gathered during system usage.

From a First Time Home Buyer's Perspective

User Insight

We wanted to get a better understanding of the background and the mindset experienced recent buyers have experienced during the current housing market. To record insight from our primary users, we conducted 7 structured user interviews targeted at new home buyers or owners with no specific age requirement spread out over a period of 5 weeks.

Below are the highlights from an affinity diagramming session done on Miro. Common themes found during this process include the real estate agent, financial remarks, search criteria, listing sites and process issues

Crowdsourcing Data

In addition to user interviews we posted a Google survey on Reddit under multiple home buying subreddits to gather further data on the home buying process, home buying tools and their demographic. From the 34 responses, some of the highlights include…

• 42.9% of home buyers were using Redfin.com to purchase a home
• 61.1% of users said searching for a home was the most difficult part of their experience
• 82.9% of user’s stated that using the home buying tools on popular sites like Zillow and Redfin helped when making a buying decision

Systems Block Diagram

Recommendation System High-level Design

Using the research we gathered we decided to best emulate the home buying process we were to divide our system into 2 sections, pre-buying and searching systems. The pre-buying phase is where we would collect user financial information (income, debts, loans, expenses etc.) and preferences (price, location, bedrooms, bathrooms etc.)

Using this information we could fetch a few preliminary listings from the Realtor.ca API based on existing criteria. Next user interaction from viewing listings, scrolling and favoriting actions would be recorded.

In the backend our system used TF-IDF to find keyword similarities between listing description and Google’s VGG-16 Pre-trained network to find image similarity between home styles. With this method we can now recommend homes based off of an averaged “ideal home” curated by continually evolving user interaction.

Design Iterations and Adjustments

Initial Ideation and Flow

After some research into other personalized experience apps I began sketching my ideas and variations of the user flow

Proposed User Flow

After a few rounds of internal feedback from the team, I arrived at the 2nd iteration site map seen below. To the team, logically this flow made sense according to our proposed algorithm. Since we haven’t done any user testing we don’t actually know if this is the best answer to resolve our product or design goals.

Design Flaws

Real-world Prototype Feedback

To test this iteration we conducted 3 user tests which included a cognitive walkthrough, post interview questions and a standard usability score test. User’s were asked to fill out the onboarding flow and browse through the recommendations within our Figma prototype.

In reality the feedback we received on our prototype was less than stellar. Generalized assumptions we made based on our user interviews were incorrect and revealed the various flaws within our prototype.

Data Privacy Awareness

One issue users immediately raised were concerns over financial and personal data that was benign collected. All users took the time to ask questions about data privacy and storage

To combat this we wanted to be transparent and directly address why we were collecting their personal data as well as what ISO standards were satisfied storing their data.

Detailed Search Results

Another issue was that users were frustrated with the lack of listing details per search result. From our initial search results, users lacked the context on property location or housing details which were buried within each listing

As an improvement we listened directly to user concerns and increased the quantity of data being presented to the user. This included property maps, search filters used, bedrooms, bathrooms, square feet and year built.

Increased Listing Visualization

Similarly, users found reading listing details difficult and would have appreciated a map to understand relative location or general neighborhood

To keep design language consistent, we made similar updates to the listing details like we did to the search results page. We primarily focused on readability and displaying the information in a way that was easy to digest.

Establishing a unique brand Identity

Keeping it Simple

Our original plan was to not have any logo at all and just “Home Simple” but our prototype felt incomplete without one. The motivation behind our logo was to keep things simple and to let users know we’re in the property technology space.

Color and Typography

On the second iteration of Home Simple it occurred to me that our blue and white branding colors were not unique to our product. Other popular realtor sites such as Zillow or Redfin did not incorporate a balanced dark theme that I felt would standout. Leading to the introducing a cool-toned sage green and accent pastel pink was done to reinforce the growth of one’s home.

Final Solution

Feature Highlight

Below is the final iteration of wireframes that our capstone project ended on. In addition, I was also able to add motion to my prototypes using protopie and bring the product design to life. Although we had to cut a few features, I’m ecstatic to see how much progress we’ve made looking back to our project start date

Deconstructive Homepage

A modern and simple home page, easy to develop and breaks down the main reasons why users should use our product

Redesigned Navigation

A revamped user interface to give user’s more detail and get a better grasp on housing details

Favorite Recommendations

Recommendation algorithm shows listings based on visual and property listings that users have liked or viewed

Final Results

Data Analysis and Interpretation

Through multiple user tests, we conducted a system usability scale (SUS) analysis with each iteration to track usability improvements in conjunction with our recommendation algorithm improvements for n = 3. We realize that a small sample isn’t representative of our real world product performance, therefore our design improvement milestones will require further testing to be certain.

• Mean user SUS score increased from 51.6 to 65.83
• Our Min SUS score floor was increased from 10.03 to 57.05
• Our Min SUS Grade rating increased from an F to a D
• While our margin of error was corrected from 50.63 to 8.77

Retro & Reflection

Documentation is everything

Part of our Capstone project requirements was to document key findings, mistakes, our process, test results and project management related activities into some form of an engineering log book. Initially it was a hassle and a task that I only did because it was a requirement by the professor.

Flash forward 8 months to the end of our project and I’m so glad I kept a record of all the research and rationale behind the decisions we made as a group. It was extremely handy to go back to view a timeline of our progress and a complete log of all our testing outcomes. You never realize how important documentation is until you need it.

Engineering Limitations

As a team it was important to identify our strengths and weaknesses to define a scope that was still feasible while balancing other courses. Early on we realized that core technology such as the image recognition library was readily available while ideas to implement housing databases or neighborhood analytics would have been nice but locked behind a paywall

Design features such as social media registration or integrated neighborhood analytics would have greatly enhanced the user experience but considered unnecessary engineering work in the context of this project

Users are in the Driver Seat

If there’s one thing we all learned early on in the engineering design process was to establish a repeatable, robust and accurate testing protocol that could be conducted after every iteration.

This way we were able to make hypotheses and validate whether or not the decisions we made were better or worse in comparison to our initial hypothesis. This came in the form of user testing, SUS scores and One-way ANOVA Statistical test.

End of a 5 Year Journey

Technically this capstone represents the cumulative application of all our knowledge, co-op experience and engineering expertise that we’ve learned in the past 5 years. Overall I am extremely proud of what we’ve created and think back to 8 months ago when we were all on a Zoom call brainstorming about which direction to take our project.

Although there were some tight deadlines to hand in deliverables and I personally struggled at times learning about how to program the image recognition software, this project has helped me grow tremendously technically and design wise.