Herman Chan· Acquisitions & AI tooling
Currently Shipping AI into a real-estate acquisitions workflow Location San Mateo, CA Updated 2026-10-20
Intro

I build AI and data systems for real estate investment.

I’m a real estate acquisitions director turning hands-on experience into working software. These three systems prepare deal analyses, track upcoming housing supply and compare rental pricing. I define the workflows, connect the data and verify the results.

Underwriting agent

Why I built it

Every deal requires similar preparation: pull property facts from the offering memo, organize the financial statements, analyze the rent roll and assemble inputs for the model.

How it works

The pack brings together the deal summary, operating analyses, rent comps, supply, demand and property-tax inputs, with source references and flagged differences.

What changed

Preparing a deal’s analysis used to take me about four hours. Now the agent handles the preparation, and I spend about thirty minutes reviewing the outputs, resolving discrepancies and confirming assumptions.

  1. Input

    Offering memo, T-12 and rent roll

  2. Process

    Extract, enrich, calculate and check

  3. Output

    Analysis pack and model inputs

~4 h
Previous manual preparation per deal
~30 min
Current human review per deal
100+
Deals with AI-built outputs

Supply database

Why I built it

Understanding future competition means piecing together development activity across separate government websites. Those filings can also identify builders and projects worth approaching before an acquisition opportunity is marketed.

How it works

Source-reported figures and estimates are distinguished, with coverage varying by jurisdiction.

What changed

I can research development around a property and investigate builder activity from one database. The same information supports the underwriting pack’s supply analysis and questions asked through an AI assistant.

  1. Input

    Environmental filings, permits and planning cases

  2. Process

    Collect, classify and connect

  3. Output

    Mapped projects with development stages, unit counts and builder information

9
Metros
254
Tracked source feeds
441,000+
Source records

Rent database

Why I built it

Building a rent-comp set means visiting leasing websites, comparing floor plans and checking concessions. Prices and offers change, making that research a recurring task.

How it works

Code reads supported leasing platforms, with AI handling pages that need interpretation. Each observation keeps its collection date and the advertised offer for review.

What changed

The database supplies rent comps to my underwriting pack, reducing manual collection and showing concessions alongside asking rents. I can focus on whether the properties are comparable and which rent assumption the evidence supports.

  1. Input

    Community leasing websites

  2. Process

    Read prices, availability and concessions

  3. Output

    Asking and net-effective rents for underwriting comps

22
Metros
16,000+
Communities tracked
202,000+
Historical rent records