About the Project: ChatAPT

Inspiration

Searching for an apartment has always felt broken. Most platforms (like Zillow or Apartments.com) make you endlessly filter, scroll, and cross-reference multiple sites. As a renter myself, I saw how painful it was for students, young professionals, and people relocating to cities like Chicago to find places that fit their actual lifestyle — not just price and number of bedrooms.

I wanted a search tool that acted like a smart friend: you could just say “I need a modern 1-bedroom near a gym and coffee shops, with a 45-minute train commute to Union Station” and instantly get curated matches. That was the spark for ChatAPT.

What It Does

(Only works in Chicago for now) ChatAPT is an AI-powered apartment search platform that translates natural language queries into lifestyle-driven matches. It doesn’t just pull listings — it analyzes them against multiple dimensions:

Commute time (calculations to places like workplaces or schools)
Points of interest (gyms, cafes, parks, nightlife)
Visual aesthetics (light, modern kitchens, large windows) via computer vision
Lifestyle filters (roommate matching, subleases, short-term rentals -- coming soon)

The goal is to make searching feel intuitive and personal, instead of like filling out a government form.

How I Built It

Frontend: Flutter
Backend: Python (FastAPI), powered by LLMs. Google commute apis
Search Engine: Typesense for fast search
Data Sources: AppFolio

Challenges I Ran Into

Data fragmentation: Apartment data is scattered across platforms, with inconsistent formats. Scraping + normalizing this data was a constant battle.
Speed vs. depth: Early prototypes felt too slow (agents reasoning on every step). I had to pre-compute commute grids, cache embeddings, and refine orchestration to make search “blazingly fast.”
Monetization experiments: Balancing free credits vs. Pro subscription, landlord partnerships, and badges for premium listings took iteration.
UI: Some testers preferred traditional filters, others loved pure chat. Finding the right middle ground was tricky. I'm not a designer so i struggled with making it beautiful and snappy.

What I Learned

Building AI agents isn’t enough — you need orchestration, caching, and real-world constraints to make them usable.
Gen Z and early professionals don’t just want listings, they want apartments that fit their lifestyle.
Distribution matters as much as tech: I experimented with Instagram/TikTok reels, Reddit posts, and guerrilla marketing to drive traffic.
Hackathons are about iteration speed — I learned to cut scope, focus on a core “wow moment,” and ship.