Scout

Apartment hunting is a part-time job nobody signed up for. ApartmentAgent does the calling, the screening, and the scheduling, so the only time you show up is when it's actually worth it.

Comment

Inspiration

We're all seniors at Minerva University with jobs lined up in different cities after graduation. When we started apartment hunting remotely, we hit a wall. Hours spent calling listings, days waiting for responses, only to find out the place doesn't allow pets or has hidden fees. We're juggling senior theses and job prep. We don't have time to play phone tag with landlords. But we can't wait until after graduation either, because prices spike and good places disappear.

Apartment hunting is broken. Listings don't tell you what actually matters. Pet policies, noise levels, management quality. You only find out after wasting time at a viewing. We needed something that could call for us and only send us to places worth seeing.

What it does

ApartmentAgent calls apartment listings on your behalf. You set your requirements (budget, pets, noise tolerance), and it monitors new listings, calls property managers, asks your questions, and only books viewings for apartments that pass all checks.

It makes two types of calls:

  • Pre-screen (60s): Checks dealbreakers - availability, pet policy, noise
  • Deep screen (4-6min): Asks about management quality, maintenance, pest issues, rent stability

It scores each place and auto-schedules viewings to your Google Calendar for the ones that match.

How we built it

Five-stage pipeline:

  1. Airbyte - Real-time listing monitoring with CDC
  2. LangGraph - Workflow orchestration with state persistence (Postgres/Ghost)
  3. Bland AI - Two voice agents with conditional logic that adapts to responses
  4. Auth0 - Login + Google Calendar OAuth for booking

Frontend: Next.js with dashboard, live transcripts, calendar view
Backend: FastAPI with JWT-protected endpoints
Compliance: Fair Housing Act enforcement—never asks prohibited questions

Challenges we ran into

Auth0 + Google Calendar OAuth was tricky. We needed the agent to write to users' calendars without storing credentials. State persistence was another challenge. We used LangGraph's checkpointer so the system can resume after crashes without double-calling listings.

Accomplishments that we're proud of

Built a working end-to-end system in under 8 hours. The agent makes real calls, parses responses, scores apartments, and books viewings. The UI is polished, the backend is secure, and it's multi-tenant ready.

We are proud of the Fair Housing compliance layer. The agent never asks prohibited questions and flags violations when landlords bring them up.

What we learned

How to build truly agentic systems that handle ambiguity and make autonomous decisions. How to integrate multiple APIs (Auth0, Bland AI, Airbyte, Google Calendar) into a cohesive product under time pressure.

Technically: LangGraph orchestration, Bland AI's multi-agent system, Auth0 OAuth flows.

What's next for ApartmentAgent

  • Scam detection - Score listings for credibility before calling
  • Learning loop - Agent learns from user feedback to predict "worth your time"
  • Negotiation - Ask about waived fees, flexible move-in dates, deposit reductions
  • Expand - Apply to other high-friction markets where calling is still the norm

Built With

Share this project:

Updates