I found my dream NYC apt with $20, Zapier, and Vercel

I recently moved to NYC and learned the hard way that apartment hunting here is not a search problem. It is a speed problem, and I was refreshing StreetEasy like it was a slot machine.

The good apartments vanish fast. Sometimes in a day. Sometimes in a few hours. By the time a listing hits your inbox, gets reviewed, gets researched, and gets sent to a broker, someone else may already be scheduling the tour.

I found my dream NYC apartment because I stopped doing that manually and built a $20 AI agent with Zapier and Vercel instead.

But first, I had to lose the perfect one.

A 2-bed in Brooklyn. In-unit laundry, elevator, no broker fee, close to the subway, bright, renovated. I had been refreshing StreetEasy for weeks looking for exactly this listing.

By the time I got to it, it was already two days old, and before I could book a viewing, someone else rented it.

That was the moment I stopped apartment hunting like a tenant and started thinking about it like an engineer.

If the best listings disappear in hours, refreshing StreetEasy once or twice a day is already losing. I didn't need more discipline. I needed an agent.

Why I built it

The annoying part was that I knew exactly what to do. I just had to do it over and over, fast.

Every promising apartment needed the same checklist: budget, subway distance, photos, amenities, building reputation, neighborhood safety, broker outreach. That's too much tab-switching when the market rewards the first qualified person to move.

The job for the agent was not "find me apartments." StreetEasy already finds apartments.

The job was:

Detect new listings before they got stale.
Research the context that listing pages leave out.
Decide whether the apartment matched my actual taste.
Act by putting only the best matches in front of me.

Every time a new listing showed up on my saved search, I ran the same workflow:

Open the listing. Check the price against my budget.
Paste the address into Google Maps. Is it on the right block? Where's the nearest subway? How far is the walk?
Skim photos and description. In-unit laundry? Elevator? Real photos or Photoshopped angles?
If it passed, run a building reputation check. Search Reddit for the address, the building name, and the management company. Some buildings had bedbug histories. Others had shady management.
Cross-check the NYC crime map for the block.
If everything still looked good, message the broker.

Fifteen minutes of careful work per listing.

That sounds fine until you multiply it across five neighborhoods, every day, while the best listings are aging by the hour.

The AI agent's whole job was to compress those fifteen minutes into an always-on loop and only forward me the best ones.

So what was the natural next step? I built one!

A $20/month Vercel app with the Zapier SDK wired into the apps I actually use.

How it works

The agent is simple on purpose.

Every 25 minutes, Vercel Cron wakes it up. It pulls my saved searches, filters out anything it has already seen, researches each new listing, scores it against my written preferences, and then uses the Zapier SDK to push the result into my email.

The pipeline looks like this:

Vercel Cron
-> scrape new listings
-> dedupe against Vercel KV
-> enrich data with maps, crime, Reddit, and web reputation
-> score and forward the best using Zapier SDK

In plain English:

Scrapes 5 of my favorite NYC neighborhoods every 25 minutes
Enriches each new listing with crime data, Reddit reviews, and a synthesized building reputation report
Scores each one 1–10 against a written-down version of my taste
Emails me only the perfect matches

No StreetEasy alerts. No daily refresh. No anxiety. Just a quiet inbox most days, and a clean shortlist when something good hits the market.

That is how I found the apartment.

The three moving parts

1. Scrape

Every 25 minutes, Vercel Cron wakes up the agent and Parallel Extract pulls my saved StreetEasy searches.

The agent parses the listings, checks them against a seen list in Vercel KV, and ignores anything it has already sent me. Only genuinely new apartments move forward.

2. Enrich

For each new apartment, the agent grabs the context the listing page will never give me: location, nearby serious crime, Reddit mentions, building reviews, management reputation, and anything weird people have said about the address online.

This is the part that makes it useful. I do not just want "2 bed, under budget, in Brooklyn." I want to know if the building has bedbug complaints, if the management company is a nightmare, or if the block is secretly bad.

3. Score and inform: Zapier SDK

Then the Zapier SDK takes over.

First, Claude scores the apartment through AI by Zapier, so I do not have to maintain a separate Anthropic API key just for scoring. It checks the apartment against my actual taste: neighborhoods I like, must-haves, subway distance, crime, building reputation, and deal-breakers.

It returns a 1-10 score, a short rationale, and a list of deal_flags.

Then the same SDK forwards the listings that clear the bar:

Anything scoring 7+ gets sent to me
Anything below that stays out of my inbox

That is the whole trick. The prompt is not magic. It is just my apartment-hunting brain written down once. When I change my mind, I edit the rubric and the agent gets smarter forever.

The part that impressed me most was that it did not behave like a dumb filter. A few apartments looked like 6s on the strict rubric, but Claude still pushed them to 7 because they had enough upside to be worth checking, like being unusually close to the subway.

This is what made it feel like a real agent instead of a smart report generator. It did not just find apartments. It scored them, filtered them, and put the best ones where I would actually act on them, without me maintaining a pile of API keys.

The design of the email. NOT MY APARTMENT.

The $20 part

Honest breakdown:

Vercel Pro: $20/mo for cron, hosting, and KV storage
Parallel Extract: gives you $80 free credits when you sign up
Claude scoring: <$1 by using Haiku
Zapier SDK: currently free during beta

The best part was not the price. The best part was that I could go to sleep and it kept checking apartments for me.

It took me one evening to build it. But it took two months of manual apartment hunting to know what to build. All those nights refreshing StreetEasy, checking blocks, reading Reddit threads, and learning what I actually cared about became the "market research" for the agent.

What stays human

The agent doesn't tour. It doesn't negotiate. It doesn't read the broker's energy in the lobby. It doesn't notice the laundry room is in the basement and the basement smells like cigarettes.

It removes the noise so I can spend my evenings on the shortlist, not the firehose.

That's the whole game. AI as a multiplier, not a replacement. The agent gives me back the most precious resource in NYC apartment hunting: attention.

THE END.

I'm Andrew. I'm doing AI at Zapier. If you also live in New York and want to talk about AI, Agents, or any other personal automations like this, let's meet.

And if you are interested in the apartment app, let me know. The agent is still running.