Currently Shipping AI into a real-estate acquisitions workflow Location San Mateo, CA Updated 2026-07-03
N° 001 / Writing / May 17, 2026 / 1 min read

What this site is

An introduction to the site — who I am, what I build, and what to expect here.

I’m Herman Chan. I work in acquisitions at Haven Realty Capital, buying and underwriting rental housing across the Sunbelt — single-family rentals and build-to-rent communities in Phoenix, Houston, Atlanta, Dallas, and the Tampa exurbs. The job is half deal flow, half spreadsheets. Read the broker’s pitch document, build the model, pull rent benchmarks from similar properties nearby, decide whether the numbers work. Do that long enough and you start noticing which parts are actually underwriting and which parts are just moving data from one place to another.

These days I also build AI tools to handle the second kind. I should be upfront about what “building” means here, because I don’t write the code — Claude writes the code. My job is deciding what each tool needs to do, where a plain script is the right answer and where a language model is, and then directing the build and testing the result on real deals. That turns out to be most of the work. The wins are specific: a one-page deal summary that used to take two hours now takes fifteen minutes, and the rent-benchmark and income-statement exercises that used to eat an afternoon each now run off a single prompt.

This site is where I write down the design decisions behind those tools. What was hard, what I tried first and threw away, what I’d do differently if I started over. Short posts, one idea each. Builder notes — not tutorials, and not predictions about where AI is going.

One filter for everything here: if I write about a tool, it’s one I run on live deals every week.

End · ~270 words Filed May 17, 2026
§ Continue reading All posts →
  1. 007 The dashboard was wrong for two weeks. Every health check was green. Jun 10, 2026 · AI
  2. 006 The feature that tested seven points better and changed almost nothing Jun 8, 2026 · AI
  3. 005 An accuracy number is meaningless without the cost of being wrong Jun 5, 2026 · AI