Real eLife

Describe Your Life. Find Your Home.

Comment

LifeMatch — Find Your Home, Not Just a House

Inspiration

Apartment hunting is broken. Every search starts the same way — bedrooms, bathrooms, price range, square footage. But none of that tells you whether you'll actually enjoy living there.

We kept asking ourselves: what if you could just describe your life? What if instead of filters, you could say "I bike everywhere, I love rock climbing, and I need to be close to UCI" — and the search actually understood you?

That question became realELife.

What It Does

LifeMatch is a natural language apartment search that finds homes based on how you actually live. You describe your lifestyle — your hobbies, your commute, your vibe — and LifeMatch parses what matters most to you, asks follow-up questions to fill in any gaps, and surfaces apartments scored against your real priorities.

Each result comes with a lifestyle match score and a visual map showing your apartment alongside the places that matter to your daily life — gyms, coffee shops, your campus, your workplace.

How We Built It

  • Frontend: SvelteKit — fast, lightweight, and great for building reactive UIs quickly under hackathon time pressure
  • AI Layer: Gemini 2.5 via the AI SDK and OpenRouter to parse natural language lifestyle descriptions into structured JSON, which we use to score and filter listings
  • Apartment Data: Stored and served locally, with attributes that map to lifestyle factors
  • Heatmaps: Generated heatmaps for lifestyle dimensions like nightlife intensity, walkability, and neighborhood quiet — giving users a visual sense of each area beyond just the listing

The core technical challenge was the translation layer: taking something as fuzzy as "I like a chill vibe and need to be bikeable to campus" and converting it into structured data we could actually compute against. Gemini 2.5 handles this beautifully — we prompt it to return clean JSON with weighted lifestyle priorities extracted from the user's message.

Challenges

Structuring unstructured intent. Human language is messy. People describe the same preference in a hundred different ways. Getting the LLM to reliably extract structured, weighted priorities from freeform input — and doing it consistently — took significant prompt engineering.

Scoring apartments meaningfully. A match score is only useful if it actually reflects what the user cares about. We had to build a weighted scoring system that respects priority ordering — your commute to school matters more than proximity to a coffee shop — and surfaces that reasoning transparently to the user.

Balancing speed and richness. The LLM parsing step introduces ~10 seconds of latency. We designed around this by making that moment feel intentional — showing the user what the AI is doing (reading your lifestyle, extracting priorities, structuring your search) rather than just a loading spinner.

What We Learned

Natural language is a genuinely better interface for high-context decisions. Filters work when you know exactly what you want. But apartment hunting is a lifestyle decision — and lifestyle doesn't fit in a dropdown. Building LifeMatch convinced us that for the right problem, removing the UI is the UX.

Built With

Share this project:

Updates