FlatFinder

Inspiration

Finding an apartment in NYC can feel like searching for a needle in a haystack. Listings appear and disappear within minutes, across multiple platforms, often with inconsistent information. We wanted to create an AI-powered personal apartment scout that automatically searches multiple platforms, filters listings based on a user’s preferences, and notifies them instantly — all in real-time. The idea was to turn the stressful, time-consuming apartment hunt into a smooth, intelligent, and even fun experience, with multi-language support and AI-generated tour videos.

What it does

FlatFinder is your loyal AI apartment retriever. Users enter their preferences — location, price range, bedrooms, amenities, and language — and FlatFinder does the rest:

Three web-scraping agents continuously monitor StreetEasy, Zillow, and Craigslist for new listings.

A manager agent consolidates results, removing duplicates and choosing the most complete or cheapest listing.

New entries trigger an LLM-based analysis to match listings to user preferences.

If a match is found, the system sends an SMS notification with a link to the listing, automatically translated into the user’s preferred language.

Each SMS also includes an AI-generated apartment tour image using Freepik.

In short: FlatFinder finds, filters, translates, and presents apartments — all automatically.

How we built it

Agents: Implemented three scraping agents (StreetEasy, Zillow, Craigslist) that gather listings in real time.

Manager Agent: Consolidates duplicates, chooses listings with the most info or lowest price, and stores them in a database.

Matching & Notification: New listings trigger an LLM to check for matches to user preferences. Matches are sent via SMS (Twilio) and translated using DeepL.

AI-Generated Tours: Freepik API generates visual previews of the apartments.

Orchestration: Airia or a simple Python scheduler coordinates scraping, consolidation, matching, and notification.

Database: SQLite or MongoDB stores listings and prevents duplicate notifications.

Challenges we ran into

Data inconsistencies: Different websites have inconsistent formats, missing fields, or duplicate listings.

Real-time notifications: Ensuring new listings were detected quickly and reliably triggered notifications.

Multi-language translation: Maintaining context and readability when translating listings into multiple languages.

AI-generated visuals: Making Freepik outputs relevant and visually appealing for the apartment tour.

Accomplishments that we're proud of

Successfully implemented three scraping agents feeding into a manager agent in real-time.

Built a fully automated workflow: scrape → consolidate → match → translate → notify.

Integrated SMS notifications, translations, and AI-generated apartment tours in a single MVP.

Created a demo-ready system that can simulate new listings appearing and instantly notify users.

Designed a product with scalability and user personalization at its core.

What we learned

How to coordinate multiple autonomous agents for scraping, consolidation, and filtering.

Strategies for deduplication and data consolidation across multiple sources.

How to use LLMs for preference matching and Freepik API for visual content generation.

The importance of real-time monitoring and notifications in user-facing automation.

Insights into building a hackathon MVP that is both functional and demo-friendly within tight time constraints.

What's next for FlatFinder

Expand to more rental platforms and international listings.

Improve matching intelligence using ML-based recommendation models.

Add full natural-language user queries (“Find me a sunny 2BR under $3,000 in Queens with a washer/dryer”).

Introduce richer AI-generated apartment tours, potentially interactive 3D visualizations.

Build user accounts and persistent preferences, enabling notifications for multiple users simultaneously.

Built With

• structify