The context
Indian real estate runs on WhatsApp. A mid-size brokerage fields a steady stream of brochure requests, price questions and site-visit asks — and a large share land after hours, in Hindi or Hinglish, often as voice notes. Speed decides who wins: research on online leads found that answering within five minutes makes a team about 21× more likely to qualify a lead than waiting thirty. Generic automation tools break on exactly the things that matter here: Hindi voice, document permissions, and how buyers actually behave.
The problem, precisely
Leads go cold overnight. Inquiries arrive after hours; with buyers contacting several brokers, the slowest reply simply loses the deal.
Language gap. Most Indian buyers prefer their own language online, and Indian-language users are far more likely to engage when you meet them in it. Few tools handle Hinglish; almost none handle Hindi voice notes.
Manual document grind. "Send the brochure" repeated all day is hours of a salesperson's time spent copy-pasting files.
No system of record. On personal WhatsApp, when a salesperson leaves, the contacts and history leave with them.
What we built ✓ verified in code
Multilingual, voice-first replies
Detects Hindi / English / Hinglish per message and replies in kind — including transcribing and answering Hindi voice notes.
Right document, right buyer
A 3-tier permission model (builder / project / unit) means a unit's sale agreement only ever reaches that unit's owner — delivered in seconds.
BANT lead scoring + hot-lead alerts
Every meaningful turn is scored on budget, authority, need and timeframe. Cross the threshold and the broker is pinged with a summary.
Inline site-visit booking
Offers slots, confirms and sends reminders — all inside the same WhatsApp thread, no external calendar.
How it works
- 1
Message in. A webhook receives it; language is detected and cached for the conversation.
- 2
Agent decides. Loaded with history and the buyer's available documents, the tool-calling agent picks exactly one tool — reply, send_document, or flag_handoff.
- 3
Act safely. A per-conversation lock prevents double-replies; below a confidence threshold it hands to a human instead of guessing.
- 4
Multi-tenant by design. One endpoint serves many businesses; each tenant's data and WhatsApp credentials are isolated.
The outcome
The agent is built to absorb the routine load that drowns a sales team — the "send the brochure", "what's the price", "can I visit Saturday" volume — answering instantly, in the buyer's language, around the clock, so salespeople spend their time only on genuinely qualified, ready buyers. Because it's multi-tenant and credential-isolated, a new brokerage can be onboarded onto its own WhatsApp number without touching anyone else's data.
The stack ✓ verified in code
Sources
- 1. WhatsApp users in India (~535.8M, 2025) — Statista. www.statista.com
- 2. Lead-response research (within 5 min vs 30 min) — MIT / Oldroyd study via Harvard Business Review, 'The Short Life of Online Sales Leads.' hbr.org
- 3. KPMG–Google, 'Indian Languages: Defining India's Internet.' assets.kpmg.com