The Problem
Real estate agents are spread thin. A single agent might be responsible for multiple listings simultaneously — phone calls, client meetings, paperwork, and physical showings all competing for the same hours.
The result? Buyers arrive at a property and wait. Or they don't arrive at all because scheduling is too much friction. Every missed showing is a potential deal lost — not because the property is wrong, but because the system failed to support the buyer at the moment of highest intent.
The Old Way
Agent must be present. Scheduling friction. No support if agent is unavailable or running late. Buyer experience is inconsistent and entirely agent-dependent.
The AI Way
Lock triggers AI. AI calls buyer within 60 seconds. Greets by name, guides room-by-room, answers questions in real time. Consistent, warm, available 24/7.
The Product Experience
The experience was designed to feel warm and personal from first contact — not robotic, not transactional. The goal was never just information delivery. It was making people feel genuinely cared for and respected at every step of the property journey.
Buyer Journey
For the open house event, a QR code trigger was added so guests could initiate a call manually — adapting the system to a crowded, social environment while keeping the same quality of conversational experience.
System Architecture
The system is built on a simple, event-driven principle: the door unlock is the trigger. Everything flows from that single event — no manual input, no scheduling, no human in the loop required.
Two conversational personas were built, each trained on the full property knowledge base. The routing decision was handled automatically, with both agents designed to feel distinctly human — different cadences, different warmth profiles, both optimised for the specific property context.
| Component | Detail |
|---|---|
| Trigger mechanism | Smart lock webhook → real-time event dispatch |
| Call initiation window | < 60 seconds from door unlock |
| Conversational personas | 2 — distinct voices & cadences, same knowledge base |
| Knowledge base | Full property data: specs, features, neighbourhood, pricing |
| Open house trigger | QR code (manual call initiation for event context) |
| Avg. interaction duration | ~4 minutes |
| Questions handled per tour | 6–8 real-time |
Hear It Live
These are real recordings from the live open house event on 14 Nov 2025. Two conversational AI personas — different voices, same property knowledge, same standard of warmth. These were recorded in the live environment, with background noise, real guests, real questions.
Build Snapshot
Seven days. Concept to live deployment. This wasn't a prototype — it was a production system running at a real event with 60+ attendees and zero handholding once live.
⚡ Concept → Live in 7 days · Including testing| Metric | Value |
|---|---|
| Total build time | 7 days (including testing) |
| Conversational personas | 2 |
| Real-world accuracy | 99% |
| Average interaction time | ~4 minutes |
| Typical questions per tour | 6–8 |
| Launch event | 14 Nov 2025 |
| Live attendees at launch | 60+ |
Launch — Open House
The system went live at a real open house — not a controlled demo, not a staging environment. The room was loud. Conversations overlapped. People were moving around. These are exactly the conditions that expose weaknesses in voice AI.
The agents held up. 99% accuracy across every interaction. Guests didn't just use the product — they enjoyed it. The experience felt novel and personal, not robotic.
The Environment
60+ guests simultaneously. Background noise. Overlapping conversations. People moving between rooms mid-sentence. Real-world chaos — not a lab.
The Result
99% accuracy. Zero system failures. Guests engaged, asked questions, stayed on the call. The AI felt warm and capable — not a gimmick.
Client Outcome
Organic Referral. Zero Outbound.
The client referred Iro AI to another founder in his network — without being asked. An unsolicited LinkedIn recommendation followed within days of the launch going live.
"The client referred Iro AI to another founder in his network — without being asked."
Unsolicited referral within days of going live · Followed by a LinkedIn recommendation
"An unsolicited LinkedIn recommendation followed within days of the launch."
This is the clearest validation: a client so satisfied they became an advocate before the engagement was even fully wrapped.
This is the clearest validation signal available pre-revenue: a client so satisfied by the work that they actively advocate for the product to their professional network, unprompted, within the first week of going live.
What This Demonstrates
This project sits at the intersection of conversational AI design, event-driven system architecture, and rapid product delivery. Each skill is validated not in theory but in a genuinely challenging live environment.
Conversational UX Design
Building AI agents people actually want to talk to — warm, natural, not robotic.
Event-Triggered Automation
Webhook-based real-time orchestration from physical trigger to AI response in <60s.
Voice AI Orchestration
Multi-persona routing with full property knowledge base integration.
Rapid Product Deployment
Concept to live production system in 7 days, including testing.
Real-World Resilience
99% accuracy in a noisy live environment — not a controlled lab setting.
System Architecture
End-to-end event-driven architecture from smart lock to AI conversation.
Key Learnings
Warmth is a product feature, not a nice-to-have
Guests didn't just tolerate the AI — they enjoyed it. The conversational design specifically focused on making people feel cared for, not processed. That's a deliberate product choice, not a side effect.
Live events are the ultimate stress test
No controlled demo environment catches what a real open house with 60+ people will expose. Deploying into chaos proved the system's resilience more convincingly than any internal QA process could.
Speed of deployment unlocks real signal
7 days. The value of moving fast isn't just efficiency — it's that you get real feedback, real results, and real referrals before slower-moving competitors have finished their planning cycle.
Unsolicited referrals are the real product-market fit signal
NPS, surveys, and reviews are proxies. A client who actively refers you to their network — before you've even asked — is the real thing. That's the signal this product generated within days.