Cubi Estate

Inspiration

I’m a data engineer, and I’ve spent a lot of time dealing with one of the internet’s most chaotic corners: real-estate data. Listings are spread across many websites, written in multiple languages, filled with inconsistent descriptions, duplicated ideas, and lots of useful details trapped inside messy text.

That frustration became the inspiration for Cubi Estate. I wanted to build something that could turn the real-estate big-data mess of the web into a search experience that feels conversational, structured, and actually enjoyable. And because I did not want all that data work to end in just another boring results page, I started imagining a more playful way to interact with properties through the Cubi world.

What it does

Cubi Estate is an AI-powered property search app that helps people discover homes through natural-language conversation instead of rigid filters alone. It can search, filter, compare, summarize, and explain listings while working across multiple languages.

Under the hood, Cubi turns messy listing descriptions into normalized features, so things like sea views, gated communities, solar panels, spas, or schools nearby become meaningful and searchable instead of getting lost in random ad copy. The Cubi world layer builds on top of that by making those property traits feel more visual, playful, and memorable.

How we built it

Cubi Estate uses a Next.js frontend and a FastAPI backend, with AI powering natural-language understanding and conversational search. The system processes listings from many sources and converts free-form property descriptions into a structured concept layer that supports better ranking, filtering, and explanations.

We also built supporting flows around the search itself, including compare, similar listings, summaries, reports, favorites, area comparison, and multilingual interaction. The playful Cubi world idea came from using that structured concept data not just for search accuracy, but also for a more fun user-facing demo experience.

Challenges we ran into

The hardest part was not making a chat box — it was taming the real-estate data chaos behind it. Different providers describe the same thing in different ways, different countries use different vocabulary, and many important property details are buried inside inconsistent listing text.

Another challenge was balancing seriousness with fun. Cubi needs to be genuinely useful as a property search engine, but the demo also needs personality, so the Cubi world had to feel playful without becoming gimmicky.

Accomplishments that we're proud of

We’re proud that Cubi Estate already goes beyond a simple property listing UI and works as a real AI-powered search experience with multilingual support, conversational search, feature-based understanding, compare flows, summaries, reports, and personalized discovery patterns. It is not just a wrapper around listings; it is an effort to make messy property data understandable and useful.

We’re also proud of the product direction behind the Cubi world idea. After spending so much effort structuring the chaos of the web, turning that structure into something fun and visually engaging feels like the right kind of payoff.

What we learned

We learned that in real estate, the biggest problem is often not lack of data, but lack of structure. Once listing information is normalized well, it becomes much easier to build better search, clearer explanations, and more engaging user experiences on top.

We also learned that utility and delight do not have to compete. A product can solve a serious data problem while still feeling playful and memorable for users.

What's next for Cubi Estate

The next step is to push Cubi further as both a serious search engine and a more distinctive user experience. That means improving the property intelligence layer, expanding feature quality and coverage, and developing the Cubi world interaction so browsing properties feels less like database work and more like discovery.

The long-term vision is simple: turn the messy real-estate web into something structured, conversational, and fun to explore.

Built With

• fastapi
• next.js
• openai
• pgvector
• postgresql
• python
• react
• typescript