Selected work

An AI desk for brokerage.

AI systemsIn development

A reply is easy to generate. Knowing what belongs in that reply takes more work.

The problem

A broker's context is spread across email, calendar events, property records, and prior conversations. A fluent draft can still refer to the wrong property, repeat an answered question, or overlook a promised follow-up.

What I’m building

I'm building workflows that bring those records together before asking a model to help. The system keeps track of the business motion, identifies the evidence available, and prepares the next step for review. The human remains responsible for the relationship and the final communication.

The work includes reply classification, calendar coordination, deal records, and reusable operating playbooks. The useful unit is an entire workflow: what arrived, what it means, what to do, and how to confirm it happened.

A reply, examined.

Read what was actually said.

Owner’s message

“We might consider selling next year. Send me a note in October.”

No asking price. No current instruction to sell.

An email is evidence. It is not the whole record.

Illustrated workflow · Fictional example

Working on a similar problem?

Let’s compare notes