Real state OS

"The operating system for real estate agencies."

Comment

Real Estate OS (REOS) — Hackathon Submission

Inspiration

Real estate agencies today juggle leads from WhatsApp, Facebook ads, and website forms — all tracked in a messy Google Sheet. Leads get forgotten. Site visits get booked and never followed up on. Managers find out about closed deals days late. We didn't want to build "just another CRM." We wanted to build one system an agency lives in all day, where AI does the boring, repetitive work and humans focus on actually closing deals.

What it does

REOS is an AI-powered system for real estate agencies. Here's the flow:

  • A lead messages on WhatsApp → an AI agent replies instantly, asks a few questions, and scores the lead.
  • A second AI agent assigns the lead to the right salesperson based on workload.
  • A third AI agent watches for leads that haven't been followed up on and sends a reminder before they go cold.
  • A fourth AI agent books the site visit and sends reminders automatically.

Every AI action is logged. Anything that goes to a customer waits for a human to approve it first with one click — so the team stays in control while the busywork disappears. Everything shows up in one dashboard: leads, properties, conversations, tasks, and a live feed of what the AI just did.

How we built it

We split the system into three parts that only talk to each other in one clear way:

  • Backend (Node.js + PostgreSQL + Prisma) — holds all the real data and rules. It's the single source of truth.
  • Frontend (Next.js + Tailwind) — the dashboard people actually use: live inbox, lead profiles, calendar, and a log showing what each AI agent did.
  • AI Agents (built in n8n) — each agent has one job: qualify a lead, assign it, follow up, or book a visit. Above them sits an AI Orchestrator — one entry point that decides which agent should handle each event. It checks simple rules first, and only asks AI to decide when something doesn't match a known pattern.

Challenges we ran into

The hardest part was deciding who "owns" what. We almost let n8n handle business logic directly, but that would've made our data unreliable every time a workflow changed. So we drew a hard line: the backend owns and checks all data, n8n only talks to the outside world and calls AI models, and every AI result comes back through one standard format so nothing touches the database without being logged. We also had to slow down our first instinct to run everything through AI — instead, we made the orchestrator check simple rules first and only fall back to AI when needed, to keep things fast and cheap.

Accomplishments that we're proud of

  • Adding a new AI agent later needs zero backend code changes — just a new workflow and one entry in a settings table.
  • Nothing gets sent to a customer without a human approving it first.
  • Every AI decision is logged: what triggered it, what it decided, and whether a person approved it.

What we learned

The real challenge wasn't "add more AI." It was figuring out which four automations actually break the business if missing — lead reply, assignment, follow-up, and visit booking — versus which ones can wait. Focusing on just those four instead of trying to build all 13 possible AI agents at once is what made this doable in a hackathon.

What's next for Real Estate OS

Next up: a No-show Recovery agent and a CRM Update agent, once we have real usage data. After that: document management with e-signatures, automatic commission tracking for finance, a customer-facing portal, and eventually a marketplace where other agencies can share their own workflows and AI agents built on REOS.

Built With

Share this project:

Updates