HomeMe, by Master of Information and Data Science alums Stanley Yin, Ian Vaimberg, Catherine Liao, Subhasis Das, and Jane Su, aims to help Singaporeans find housing with an agentic question and answer (QA) system empowering non-technical users to easily query housing policies and find their dream homes. The team was awarded the Hal R. Varian Capstone Award for Summer 2025.
To learn more, we interviewed the team —
What inspired your project?
Stanley: Inspiration for this project began back in 2022 when my wife and I first decided to buy a public housing apartment in Singapore. In Singapore, 80% of the population lives in public housing, which, unlike other countries, covers a broad range of income groups, including the middle class. Without the public housing scheme, the costs of living would be unaffordable for the general population. At that point, we had been renting for nearly a year, and we had outgrown the tiny and very expensive apartment that we were renting, especially with the birth of our first child.
In many ways, our delayed decision to buy a flat was a result of the many complex policies and processes involved with buying public housing in Singapore. Despite having worked at the Housing and Development Board (HDB), the agency overseeing all public housing matters in Singapore, for nearly 4 years, even I had difficulty navigating the complex policies and options surrounding public housing eligibility, grants, and loan options.
It was only with the help of a family friend, who worked as a real estate agent, that we managed to work through the challenges that our family faced. Since then, I’ve always wondered if there could have been a better solution for Singaporeans, who come from many different backgrounds and contexts, to more easily assess their housing and grant eligibility requirements — or just to find more information on public housing schemes and policies in general. Conversations with many friends and co-workers reinforced my observations — this was a problem faced by many people, and something ought to be done about it! Then, having done DATASCI 221 and DATASCI 267, I realized that a RAG (Retrieval Augmented Generation)-based architecture might just be the trick for this problem that has long affected Singaporeans trying to buy a home.
“The combination of diverse perspectives and digital collaboration allowed us to produce stronger work as a team than we could have individually.”
What was the timeline or process like from concept to final project?
Stanley: We had three benefits that really enabled our team to deliver so much with so little time. First, we had a very clear picture of what features our product would have and what needed to be done to achieve them.
Second, our team had a strong talent mix — we had individuals skilled in NLP, GenAI, and software engineering.
Third, we had the benefit of a clear project management plan up front. We split our tasks into four tracks for each line of work (e.g., QA development, metrics, agent development, and ML engineering) such that each of these tracks could run in parallel. Each of these tracks consisted of 2-week sprints per task. We identified clear dependencies, the critical path, and potential choke points early on — keeping a very close watch to make sure that any dependencies were not delayed.
One good decision we made was to create a bare-bones QA agent early on, so that backend and frontend work could begin right from the first sprint! Looking back, this decision probably saved our entire timeline when unexpected delays (metrics design and set-up dragged on for nearly 6 weeks instead of 2) occurred!
How did you work as a team? How did you work together as members of an online degree program?
Catherine: Our team worked across multiple time zones, so consistent communication was essential. We scheduled weekly Zoom meetings to review progress and also used extra time after class to share updates. Additionally, we stayed connected on Slack, providing regular updates on completed tasks and addressing any questions. Tasks were assigned based on each member’s interests and skills, and grouping people with similar task interests helped foster collaboration. While we maintained autonomy over our individual tasks, we could always request help and support when needed. Overall, maintaining frequent and clear communication was important for us.
A benefit of being part of an online degree program was the ability to work with classmates from diverse geographic, professional, and cultural backgrounds. This not only broadened our perspectives but also helped us adapt to digital collaboration tools that mirror real-world remote teamwork. Ultimately, the combination of diverse perspectives and digital collaboration allowed us to produce stronger work as a team than we could have individually.
How did your I School curriculum help prepare you for this project?
Ian: The breadth of this project required the team to draw on tools and concepts from across the MIDS curriculum, with several courses proving especially influential. DATASCI 201 was pivotal as it marked the first experience working in formal data science teams within the program. This helped establish a foundation for effective collaboration on complex projects, an important shift for those coming from less cross-functional settings or with limited experience working alongside other data professionals. DATASCI 205 (Data Engineering) provided essential skills in data wrangling and introduced containerization, both of which were critical for building and scaling the infrastructure behind the project.
Most significantly, DATASCI 267 (Gen AI) enabled the team to bring the project to life by offering deep exposure to LLMs along with practical guidance in constructing a conventional RAG system. This experience served as the baseline, allowing the group to rapidly reach new levels of sophistication and innovation with a custom RAG build. The course also introduced the core ideas behind agents and the supporting tools, which guided the team as it extended the project into more advanced development. Collectively, these courses equipped the team with the technical expertise, collaborative practices, and problem-solving frameworks that directly shaped the success of the project. We would strongly recommend Gen AI, DATASCI 267, in particular, given its relevance and value in light of the rapid advances in the field.
Do you have any future plans for the project?
Jane: Our final project meeting was scheduled for just thirty minutes. Instead, it stretched to more than two hours, carrying me past midnight in New York. No one wanted to hang up. Perhaps deep down, we all knew this would be our last call as a full team, and none of us were ready to let go. Over the summer, five people who began as strangers had become collaborators, friends, and co-builders of something beautiful. In just three months, we not only shared our skills but also created a product that surprised even ourselves in its scope and polish.
The conversation that night began innocently enough. Our team lead, Stanley, asked, “What’s your plan after the class ends? I’ll go for a long walk.” From there, he painted vivid stories of Singapore: the tropical rainforest weather, the rhythm of the monsoons, the dazzling Rain Vortex waterfall at Changi Airport, and the surprising richness of the airport’s food courts. By the time we logged off, everyone was starving and already serious about meeting Stanley in Singapore someday.
That call felt less like an ending and more like a promise. Even a month after graduation, our team still checks in, reminding us that what we built together is more than an agent website but a shared memory.
But if this question is asking the next step for HomeMe, the formal answer will be to continuously unlock under areas of public housing. We only covered one of six website domains, so one clear next step is to scale the system for the others (e.g., potentially one QA agent each, with an orchestrator balancing between them).
We will also further experiment with HTML RAG architectures and configurations, tighten up existing metrics, and develop more as the app grows (e.g., guardrails, user goal achievement, etc.).
How could this project make an impact, or, who will it serve?
Subhasis: HomeMe supports a wide range of users:
- First-time homebuyers navigating Singapore’s complex HDB (Housing and Development Board) system.
- Existing HDB owners looking to upgrade or downsize through resale.
- Non-technical users who struggle with HDB’s difficult website navigation.
- Young couples, families, and singles exploring their eligibility for HDB purchase.
The Impact HomeMe Will Make:
- Democratizing Access to Housing Information: 80% of Singapore's population (3 million people) lives in HDB flats, yet the housing system is complex to navigate. HomeMe provides an intuitive AI interface, making housing information accessible to everyone. Free public access removes financial barriers to housing information.
- Reducing Financial Burden on Home Buyers: Real estate agents typically charge 1-2% of the property price. HomeMe provides expert-level guidance for free, potentially saving buyers thousands of dollars per transaction.
- Improving Decision-Making Quality: With 74,000 attempts to buy new flats annually, but only 28,000 successful transactions, many buyers make uninformed decisions. HomeMe's comprehensive database helps users make better-informed choices.
- Streamlining Complex Processes: Addresses HDB website limitations (complicated navigation, limited search filters, unclear eligibility) through a conversational interface that understands natural language queries.
Our Mission: “We’re here to create a stress-free experience for all Singaporeans looking for a public home by democratizing access to housing information.”
