Pinned Tweet
Just hit 3,000 followers!
Can't believe we're here — 3,000 of you following my journey as an AI Engineer & MCP Developer.Building with LangChain, RAG, LLMs & AI Agents every day. Grateful for every like, share & connection.
More builds. More content. More value. Let's go!
#AI #MCP #LLM #BuildInPublic #AIEngineer #3KFollowers
Proptech demos will matter less than response time. A Dubai agency that answers every enquiry in a minute with a useful reply beats one with a fancier dashboard.
Have AI draft the follow up for every viewing within an hour, using the notes the broker typed. The broker edits two lines and sends. Fast, and still sounds human.
Built a handoff rule for a broker chat agent. First version passed every lead to a human once they gave a budget, so brokers got flooded with tire kickers. Fix was waiting for a viewing date or a named project before handoff. Fewer chats reached brokers, but far better ones. What triggers a handoff on your team?
3 mistakes when adding AI to a broker CRM:
1. Letting it write to records without a human approving the first month
2. No log of what it said to each lead
3. No stop word so a buyer can opt out
Fix these before adding features.
Dubai investors will soon expect to see how a price estimate was made. Agencies that explain their numbers will earn more trust than ones that show a bigger number.
Dubai investors: what would make you trust an AI rental yield estimate enough to act on it, a shown list of comparable units or a human check?
Have AI summarize each viewing call or WhatsApp thread into three lines for the broker: budget, area, next step. Nobody reads a 40 message chat before calling back.
This week I'm building a rent estimate feature for an agency dashboard. Stuck on furnished versus unfurnished units, listings rarely label it properly so the model mixes them. Next step is a small classifier that reads the description text first. Anyone solved messy listing labels another way?
Dubai buyers will start by asking an AI assistant, not scrolling portals. Agencies with clean, structured listing data get recommended. The rest get skipped.
Let AI answer the 2am WhatsApp enquiry and ask just one thing, budget or timeline. Your broker wakes up to a warm lead, not a cold name.
3 things to automate first in a Dubai brokerage:
1. Instant first reply to portal and WhatsApp enquiries at any hour
2. Reminders for stale leads nobody has touched in a week
3. A weekly summary of which sources actually produced viewings
Skip the fancy stuff until these three work.
The next gap between Dubai agencies won't be who has AI, it's who has clean data. A smart agent on messy CRM records gives wrong answers faster. Agree or not?
Building a lead dedupe step for a broker CRM. Same buyer came in from a portal, Instagram and WhatsApp with three different name spellings, so the agent followed up three times. Matching on name failed. Matching on normalized phone number fixed most of it, and fuzzy name matching only as a tie breaker. Boring fix, but it stopped the triple messages.
Dubai brokers: when a lead goes quiet after the first WhatsApp reply, how long do you wait before following up, and is it a person or a template?
Run every new listing through an AI check for missing size, bedroom count and payment plan details before publishing. Fixing it in a minute beats a portal rejection later.
This week on my property matching tool: buyers describe what they want in plain words and it turns that into filters plus a ranked shortlist. Stuck on phrases like quiet area or family friendly, they mean different things to everyone. Next I'll test asking one clarifying question first.
Off plan buyers will expect instant answers on payment plans at any hour. Brokers who pair AI replies with a human closer win. Brokers who hide behind a bot lose trust.
3 things I check before trusting an AI price estimate for a Dubai unit:
1. Is it comparing the same building and similar floor, or just the area average
2. How old are the comps it used
3. Does it say when it does not know
A confident wrong number is worse than no number.
Built a WhatsApp lead qualifier for brokers. First version asked budget, area and timeline in one message and buyers just ignored it. Fix was one question per turn, with the next one depending on the last answer. Chats finished far more often in testing. Short turns beat clever prompts. Anyone else seen this?

