Sublety

Sublety: Swiping app for sublets and renters. Match by lifestyle & needs. Uses Gemini + ElevenLabs for instant voice onboarding. Just speak your profile, swipe, and move in.

Comment

Inspiration

As students, the scramble to find housing is a biannual ritual. Whether we are moving for a 4-month co-op term or looking for someone to take over a lease while we study abroad, the current solutions are archaic. We were tired of scrolling through endless, unformatted Facebook Marketplace descriptions and dodging scams.

We wanted to bring the efficiency of modern dating apps to the rental market. We realized that compatibility in housing isn't just about price and location; it's about lifestyle. Do you have pets? Do you smoke? Do you need wheelchair accessibility? We built Sublety to make finding a place (or a person) as easy as swiping right.

What it does

Sublety is a desktop application connecting sub-letters with renters.

  • Swipe to Match: Users swipe through cards of potential rentals. A "match" triggers a confetti explosion and opens a chat.
  • Smart Filtering: The app strictly filters matches based on non-negotiables: Gender preference, Accessibility needs, Smoking/Vaping habits, and Pet ownership.
  • Conversational Onboarding: Instead of filling out a tedious form, users talk to the app. An AI assistant interviews the user and instantly builds their profile.

How we built it

We prioritized a modern, lightweight, and reproducible stack.

The Core Application:

  • Frontend: We built the UI with React (via Vite) for speed and used Tauri v2 to wrap it into a lightweight native desktop executable (.exe).
  • Swipe Mechanics: We utilized react-tinder-card to handle the physics of the swiping interface.
  • State & Data: We avoided "prop-drilling" by using Zustand for global state management (like tracking the currentUser) and SWR for efficient data fetching and caching.

The Serverless Backend (Azure):

  • Hosting & Auth: We deployed to Azure Static Web Apps, which handled our hosting and provided out-of-the-box authentication via GitHub/Microsoft.
  • API: Our backend logic runs on Azure Functions (Node.js), allowing us to scale endpoints without managing servers.
  • Database: We used Azure Cosmos DB (NoSQL) to store our JSON-heavy user profiles and match data.

The AI Integration:

  • ElevenLabs provides the natural voice for our onboarding assistant.
  • Google Gemini parses the user's spoken responses into structured JSON for our database.

Challenges we ran into

  • Local Cloud Emulation: Running a desktop wrapper (Tauri) alongside a serverless backend locally was complex. We had to use @azure/static-web-apps-cli paired with concurrently to run the frontend, backend, and emulator in a single terminal command.
  • State Management: Keeping the "deck" of user cards in sync with the backend while handling rapid swipes was tricky. We used Zustand to decouple the UI state from the database logic.
  • Hallucinations vs. Structured Data: Getting Gemini to output clean, database-ready JSON without "hallucinating" extra fields required significant prompt engineering.

Accomplishments that we're proud of

  • Reproducible Environment: We set up a Nix shell at the project root. This meant every team member had the exact same dependencies and environment instantly, eliminating "it works on my machine" issues.
  • Polished UX: We focused heavily on "game feel." Using canvas-confetti for matches and react-hot-toast for notifications makes the app feel responsive and fun.
  • Serverless Architecture: We successfully built a full-stack app without provisioning a single server, relying entirely on Azure's serverless ecosystem.

What we learned

  • Tauri vs. Electron: We learned how to build cross-platform desktop apps using Tauri, which is significantly more resource-efficient than Electron because it uses the OS's native webview.
  • NoSQL Modeling: We gained experience structuring data for Cosmos DB, optimizing our schema for fast reads (swiping) rather than complex joins.
  • Prompt Engineering is Coding: We learned that writing natural language prompts for Gemini requires the same rigor and testing as writing standard code.

What's next for Sublety

  • Mobile Port: Since we used React and Tauri, porting this to a native mobile app is the next logical step.
  • ID Verification: Integrating a verification API to prevent scams.
  • Smart Contracts: Auto-generating sublet agreements based on matched terms.

Built With

Share this project:

Updates