Real Estate Agent Assistant

Main value delivered is scraping / filtering module based on criteria. This module works in real product and our submission shows the vision

Inspiration

"When I'm at a viewing and someone calls — customer lost."

That one line from a Vienna real estate agent captured what we kept hearing across 26 discovery calls and 148 prospects.

Three patterns kept repeating:

agents lose leads because they can't pick up the phone during viewings.
They burn 5-6 hours per search assignment manually copying listings into spreadsheets.
And they drown in email -- sorting inquiries, responding to the same questions, chasing follow-ups.

Three pain points. One persona. One root cause: the agent is the bottleneck.

So we built one AI assistant that handles all three — voice, search, and email — so the agent can focus on what actually closes deals: relationships, viewings, and negotiations.

This hackathon is where we show that vision as a single working demo, implementing the technical parts the way we'd build them in an ideal world.

What it does

The Real Estate Agency AI Assistant is a 3-in-1 tool built for real estate agents of Austria:

AI Voice Agent — Answers calls when the agent is at a viewing or in a meeting, pre-qualifies the caller (budget, urgency, timeline), captures contact details, and forwards a structured summary. No more lost leads.

Multi-Portal Property Scraper — Searches willhaben, ImmobilienScout24, ImmoWelt, and insolvency databases by criteria. Deduplicates, exports to Sheets, auto-refreshes weekly. 5-6 hours of manual research -> a 2-minute review.

AI Email Agent — Classifies incoming emails, drafts context-aware responses in the agency's voice, and queues them for human review.

(Techniclally Built as a coding agent extension in parallel session) -- because this tool is deeply personal, I built it the way I actually want to use it.

How we built it

Discovery-first.

26 call transcripts -> structured signal table -> 10 pain clusters -> 5 scored ICP hypotheses -> 11 mapped objection patterns. (this was done outside the hackathon, I don't include it)

Then we built only what the data demanded.

German-language voice recognition,
multi-portal scraping with deduplication,
email classification with embedding database support for the niche-specific knowledge.