HomeSwipe Agent

• 

  /home

• 

  /scout

• 

• 

Inspiration

Apartment hunting is messy. Listings often hide important details like broker fees, commute time, missing amenities, duplicate prices, and unclear red flags. We wanted to build an AI apartment scout that helps renters make faster, more informed decisions.

What it does

HomeSwipe Agent lets a renter enter a goal like “Astoria studio under 2500 near subway.” It uses Nimble to search the open web, OpenAI to extract structured listing facts, and a deterministic scoring engine to rank each apartment by rent, commute, amenities, lifestyle fit, and risk.

Users swipe like or pass. As they swipe, HomeSwipe learns preferences such as favorite neighborhoods, important amenities, commute sensitivity, and red flags they avoid. ClickHouse stores the agent workflow events for analytics.

How we built it

We built HomeSwipe Agent with Next.js, TypeScript, Tailwind CSS, Nimble, OpenAI, ClickHouse, and localStorage.

Nimble finds real web sources. OpenAI extracts listing facts into strict JSON. Our scoring engine calculates the match score. The swipe UI saves liked homes and learns preferences locally. ClickHouse logs events like web searches, sources found, listings extracted, and like/pass decisions.

Challenges we ran into

The hardest part was handling messy listing data. Rental pages can have missing prices, unclear fees, weak descriptions, or incomplete commute details.

We solved this by separating the system into layers: Nimble retrieves sources, OpenAI extracts facts, our scoring engine makes the final decision, and the UI falls back safely if an API fails.

Accomplishments that we're proud of

We built an end-to-end agent workflow: web search, structured extraction, scoring, swipe decisions, learned preferences, saved homes, and ClickHouse analytics.

We are especially proud that the score is explainable. Users can see exactly how rent, commute, amenities, lifestyle, and risk affect each listing.

What we learned

We learned that a useful AI agent is not just a chatbot. It needs reliable sources, structured data, explainable decisions, and workflow tracking.

We also learned that AI should not do everything. OpenAI extracts facts, but our own deterministic scoring engine makes the ranking auditable.

What's next for HomeSwipe Agent

Next, HomeSwipe could add verified commute APIs, map-based comparisons, stronger fraud detection, roommate matching, listing alerts, and landlord contact automation.

Built With

• clickhouse-cloud
• localstorage
• next.js
• nimble-web-search-agents
• node.js
• openai-structured-outputs
• react
• tailwind-css
• typescript