Solo Build · Next.js + OpenAI GPT-4o · Live on Vercel
Room.zip — End-to-End AI Co-Living Platform
Summary
Solo full-stack web app unifying roommate search and rental search into one flow. 31-question lifestyle survey feeds a two-GPT-4o-agent compatibility simulation; matched pairs receive joint AI-ranked U.S. rental recommendations via RentCast. 1K+ seed users, live on Vercel.
Product architecture — three core pillars
Defined the product around three core pillars: Deep Data, replacing bio-and-photo matching with a 31-point lifestyle survey; AI Synergy, using paired LLM agents to simulate both users’ compatibility before move-in; and Unified Search, recommending rentals based on both users’ combined housing preferences after a match.
Differentiation — roommate matching plus rental search
Positioned Room.zip against listing-first and roommate-first platforms by combining three layers in one workflow: 31-point lifestyle matching, multi-agent compatibility simulation, and joint AI rental recommendations. Unlike platforms that focus primarily on listings, chat access, or manual vetting, Room.zip connects roommate fit and housing search in a single end-to-end experience.
End-to-end roommate matching and rental recommendation flow
Designed an end-to-end workflow from Google OAuth and PortOne KYC onboarding to a 31-question lifestyle survey, algorithmic candidate filtering by gender, age, budget, and location, paired LLM-agent compatibility simulation, AI-persona pre-match chat, mutual matching, and joint rental recommendations using RentCast listings ranked against both users’ housing preferences.
Architecture and scale
Built the platform on Next.js and Vercel serverless hosting, with NextAuth.js Google OAuth and PortOne KYC for onboarding. Implemented an AI layer with 15 LLM-powered endpoints covering compatibility simulation, persona chat, and rental ranking. Managed data through Supabase PostgreSQL with Prisma models, integrated RentCast for rental listings with Zillow deep-links, and onboarded 1K+ seed users for early validation.
Multi-agent compatibility simulation — core differentiator
Built a multi-agent compatibility system that converts each user’s survey responses into a structured lifestyle persona across sleep, cleaning, temperature, guests, noise, food, and study habits. Paired LLM agents simulate a turn-based roommate dialogue to identify likely friction points, then aggregate the dialogue into a numerical compatibility score and written report. The same persona layer is reused for post-match rental ranking, connecting survey, match, and housing recommendation in one continuous flow.