Lead-Qualification Agent
An AI home-advisor for real estate that qualifies buyers, scores its own confidence, and escalates hot leads over WhatsApp — tuned by a three-layer evaluation suite.
At a glance: Ethereal built an AI home-buying assistant for a property developer drowning in inquiries. It chats with each buyer, figures out their budget and how serious they are, and hands the hottest ones straight to the sales team over WhatsApp — and it was fine-tuned by rehearsing against fifteen different types of buyer.
The Problem
A property developer was flooded with unqualified inquiries. Sales couldn't prioritize, hot buyers slipped away, and every conversation started from zero.
A single launch buries the sales team in inquiries, most of which will never transact.
With no way to rank intent, hot buyers wait in the same queue as tyre-kickers.
Every conversation restarts from zero — nothing the buyer already said is remembered or scored.
The Solution
We built an AI home-advisor that qualifies buyers in three phases — passive extraction, active probing, pre-handoff confirmation — scores its own confidence (LOW / MEDIUM / HIGH / ESCALATE), and detects intent (own-use, investment, NRI, gift, custom-build). Hot leads escalate to sales over WhatsApp. The agent is tuned by a three-layer evaluation suite: analytics, an LLM judge, and a 15-persona simulation.
Passive Extraction
The agent infers budget, intent, and timeline from what the buyer volunteers — without interrogating.
Active Probing
Gaps are filled with targeted questions, but only when confidence is too low to decide.
Pre-Handoff Confirm
Before escalating, the agent confirms the qualified picture back to the buyer.
Confidence Engine
Every turn is scored LOW / MEDIUM / HIGH / ESCALATE — the score decides what happens next.
buyer personas in the eval loop
Own-use, investment, NRI, gift, and custom-build buyers are routed differently from the first message.
Real conversation metrics surface where the agent hesitates or misreads intent.
A second model grades each transcript against the qualification rubric.
A simulated panel of fifteen buyer personas stress-tests the agent before anything ships.
Four-level scoring makes every qualify-or-escalate decision inspectable.
Hot leads hand off to a human over WhatsApp with full context attached.
The Outcome
Automated buyer qualification and escalation — and a system that keeps improving itself through its own evaluation loop.
Built with
07 technologiesHave an agentic ai problem like this?
Tell us what you're building — we'll tell you exactly how we'd ship it.
From the archive
Three parallel AI agents that audit websites, product flows, and user experiences from multiple perspectives at once — with video evidence.
Describe a multi-leg options strategy in plain English — the copilot constructs it, validates it against risk rules, and executes across accounts.
An agent that sanitizes enterprise change requests, scores compliance, and plugs straight into the ITSM workflow — with a full audit trail and zero PII leakage.
A self-hosted intraday trading cockpit built around one guarantee: your broker keys never leave your machine.