propcall

AI that calls your real estate leads in 30 seconds — before any human knows they exist.

Comment

Inspiration

Every real estate developer in India knows this feeling. You spend Rs 5 lakhs on Google Ads and 99acres. Leads come in. Your telecaller team calls back the next morning. The buyer has already visited a competitor's site and put down a token.

The problem isn't the product. It isn't the price. It's the six hours of silence between a buyer raising their hand and a human picking up the phone.

I kept seeing this pattern in every real estate client I worked with as an AI consultant. Rs 15,000 crore is lost every year across India's real estate sector — not to bad products or bad marketing — but to slow follow-up. Research is unambiguous: lead conversion drops 400% if the first call happens more than 5 minutes after form submission. Most teams call back in 4 to 6 hours.

I wanted to build something that made "missing a lead" structurally impossible. Not a better CRM. Not a smarter dialer. Something that fires before any human is even aware the lead exists.

What it does

PropCall Live is an AI lead qualification platform for Indian real estate developers and brokers. The moment a buyer submits an enquiry form, three things happen simultaneously — in under 30 seconds:

  1. Instant AI Lead Scoring — A weighted algorithm reads the buyer's budget fit (40%), purchase urgency (40%), and lead source quality (20%) to produce a 0–100 score with a HOT / WARM / COOL label. A buyer looking for a 3BHK in budget who needs to move immediately scores 94. A casual browser with no timeline scores 32. The sales team knows immediately who matters.

  2. Telegram Notification — The sales manager's phone gets an instant alert with the full lead profile, AI score, and a personalised Hindi follow-up message — generated on the spot based on the buyer's specific situation. Ready to copy-paste. No thinking required.

  3. AI Voice Call — Powered by Retell AI, an AI agent calls the buyer's phone in natural Hindi-English, introduces itself as calling from the developer, asks qualifying questions, and books a site visit. The buyer fills a form at 11PM. Their phone rings within 30 seconds.

The platform also includes a real-time CRM dashboard with live lead scores and call statuses, a weather API integration that recommends the best upcoming days for site visits based on live forecasts at the project location, and an interactive ROI calculator that shows exactly how much revenue a developer is leaking and what PropCall Live recovers.

How we built it

The entire platform was built using MeDo — described in plain English, iterated in minutes, deployed the same day.

The stack: React 19 + TypeScript + Tailwind CSS for the frontend, Retell AI's Web SDK for browser-based voice calling, Telegram Bot API for real-time notifications, Open-Meteo for weather data, and Vercel serverless functions as a secure proxy layer.

Every sensitive credential — Retell API key, Telegram bot token — lives exclusively in server-side Vercel environment variables and never touches the browser bundle. The client calls our own /api/create-call and /api/send-telegram endpoints, which proxy to the upstream APIs server-side. MeDo architected this security model and implemented it correctly from the start.

The lead scoring algorithm, Hindi message generation engine, localStorage persistence, weather integration, and ROI calculator — all described conversationally to MeDo, all shipped as production-quality TypeScript. What would have taken a three-person engineering team three months took days.

Challenges we ran into

CORS on Retell's API — The create-web-call endpoint is designed for server-side use. Calling it directly from the browser triggered CORS errors in production. The fix was building a Vercel serverless proxy — but identifying that this was the root cause, and not an SDK issue or a credentials issue, took real debugging time.

The broken CDN — The original implementation loaded the Retell Web SDK from a CDN URL (@retell-ai/retell-client-js-sdk@1.0.0) that returned a 404. The actual package is retell-client-js-sdk (no @retell-ai/ scope) and the global export was window.retellClientJsSdk.RetellWebClient — not window.RetellWebClient as assumed. Switching to a proper npm import resolved this entirely.

Hindi that sounds human — A static template like "Namaste {name} ji, aapka enquiry ke liye shukriya" gets repetitive fast. Building message variations that feel genuinely contextual — different urgency for "Immediately" vs "6+ Months", different openers for Walk-in vs Google Ad sources, different tone for HOT vs COOL leads — required thinking carefully about what a good human telecaller actually says in each scenario.

Making the ROI calculator honest — The follow-up speed slider initially had no effect on the calculation. Wiring it to the lost-leads formula (faster follow-up = fewer leads lost, slower = more lost, capped at 75%) made the tool actually teach something rather than just showing numbers.

Accomplishments that we're proud of The 30-second pipeline actually works end to end — form submission triggers real Telegram notifications and real Retell AI voice calls, not mocked demos.

No secrets in the browser bundle. Every API key is server-side. We checked the production build explicitly and confirmed zero credential leakage.

The Hindi message generator produces genuinely different, contextually appropriate messages based on score, timeline, and lead source — not the same template with a name swapped in.

Building a production-grade, multi-integration AI platform — with a working CRM, voice calling, real-time notifications, weather API, and a secure backend layer — without writing a single line of code manually. The entire product was built in conversation with MeDo. That still feels remarkable.

What we learned

Speed of response is the product. In Indian real estate, the value proposition isn't features or pricing or design. It's the 30 seconds. Everything else in the product exists to support that one moment.

MeDo changes what's possible for solo founders. The traditional barrier to building something like PropCall Live was engineering resources — minimum 2-3 developers, 2-3 months, Rs 15-20 lakhs in salaries. MeDo collapsed that entirely. The constraint is now clarity of thinking, not ability to execute.

Security has to be first, not last. We started with API keys hardcoded in the client bundle — which is the natural first instinct when prototyping. Moving them to server-side proxies required restructuring multiple files. Building it securely from the start would have been faster.

The meta-story is as powerful as the product. PropCall Live is an AI product. It was also built by AI. That recursive story — AI building AI for AI use cases — resonates with audiences in a way that straightforward feature pitches don't.

What's next for PropCall Live Outbound phone calling — The current implementation does browser-based web calls. The next version initiates actual outbound calls to the buyer's phone number via Retell's phone calling API, so the AI calls even when the buyer has closed the browser.

White-label for developers — A self-serve onboarding flow where any developer configures their project name, budget ranges, and agent script without touching code. Each developer gets their own branded form and dedicated Retell agent.

CRM integrations — Push scored leads directly into Kylas, Sell.Do, or a custom webhook so PropCall Live sits on top of whatever CRM the developer already uses.

Multi-language expansion — Tamil, Marathi, Telugu. Indian real estate is hyperlocal. A buyer in Chennai needs a different agent than a buyer in Noida.

Consulting-led GTM — PropCall Live is the centrepiece of an AI consulting practice for real estate developers. Every consulting engagement starts with a live demo. The demo closes the contract. The contract funds the product. This is already working.

Built With

Share this project:

Updates