Real-estate investor (USA).

The problem

An investor buying multifamily properties in the US receives offerings the way brokers send them: by email, often as marketing flyer images, each in a different format. Manually retyping parameters, checking whether "Sunset Apartments" is the same property as "Sunset Apts LLC" in the database, and tracking offer deadlines consumed time that should have gone into investment decisions.

What I built

A pipeline that, every hour, turns an inbox into a structured transaction funnel:

• Classification & extraction — AI separates deal emails from the rest, reads property parameters, broker details and deal terms from the email body and the attached flyers, and merges everything into a single verified record.
• Three-tier matching — first an exact match against the property database, then a fuzzy algorithm resilient to spelling differences, and finally an AI agent that tries successive name variations before declaring "not found" — always with a confidence level and reasoning.
• Automatic enrichment — geocoding and a map link, ratings of nearby schools, neighbourhood demographics, sale history and a computed price per unit.
• Business rules — every deal gets a deadline: off-market an automatic internal one, marketed deals the broker's date. Nothing sits in the funnel without a clock.
• Nothing disappears silently — non-deal emails, geocoding failures and unreadable offerings go to dedicated review views with a plain-language reason.
• The human decides — Letters of Intent are generated in one click from a template, offer prices and dates are edited via forms, and planned property viewings are arranged into an optimised route and emailed.

The outcome

The system ran hourly, 24/7, turning the stream of broker emails into ready, enriched investment records — and preparing an LOI went from manual drafting to picking an item from a list. After delivery and a stabilisation period, the system was handed over and now runs at the client's independently.