Claude Code for Real Estate

Material from a Claude Code workshop I ran for real estate people. The logistics and install steps have come down; the model and the prompting notes are still here.

Get the model running

Assumes you already have Claude Code installed. Paste this into Claude Code and press Enter:

Clone https://github.com/lsarta/embarcadero-model.git into a folder on my desktop, then run the dashboard and open it in my browser.

Claude will ask permission before running commands. That's a feature. Approve and keep moving.

✅ Success: your browser opens a dashboard and the big number reads 12.62%.

Try this first

In Claude Code, type:

Change the exit cap to 7% and re-run the model. What happened to the IRR?

Then click "Re-run model" in the dashboard and watch the number move.

The model

A five-year unlevered DCF for a hypothetical acquisition of Two Embarcadero Center. deal.json holds the assumptions; model.py is the model.

Metric	Value
Basis (incl. closing)	$395,850,000
Year 1 yield on cost	6.75%
Exit value	$527,053,284
Unlevered IRR	12.62%
Equity multiple	1.71x

It also ranks which assumptions actually move the answer:

Assumption	IRR swing
Annual rent growth	6.46%
In-place rent ($/SF)	6.14%
Purchase price	4.58%
Exit cap rate	3.66%
Operating expenses ($/SF)	3.44%
Stabilized occupancy	3.15%
Annual opex growth	0.85%
Months to stabilize	0.20%

Every input's range and sourcing is in deal.json. Disagree with a band? Change it.

The building is real. The deal is not. Two Embarcadero Center is owned by BXP and not for sale. Assumptions are illustrative — see SOURCES.md. Not investment advice.

How to ask for what you want

Specify like you're briefing an analyst.

Weak add a sensitivity table

Strong Add a two-way sensitivity grid to this model. Exit cap rate across the columns from 5.75% to 7.00% in 0.25% steps, annual rent growth down the rows from 0.00% to 4.00% in 0.50% steps. Each cell shows the unlevered IRR. Put the grid in the dashboard below the assumption ranking. Don't change the values in deal.json — the base case should still come out at 12.62%.

Say what you want, what "done" looks like, and what not to break.

Make it yours

The model runs. Now change it into something you'd actually use — pick something from your own work, not a generic exercise.

You can't break anything permanently. git checkout . puts the model back exactly how it started.

Don't know where to start?

Sensitivity grid. A two-way table across exit cap and rent growth.
TI/LC. This model currently doesn't account for TI and LC.
Rent roll. Individual leases with expirations instead of one blended rent PSF.
Waterfall. LP/GP split, preferred return, promote.
Monthly periods. Currently annual only.
Excel export. Write the cash flow to a spreadsheet instead of the terminal.
Downside case. A recession scenario with re-tenanting drag.

Or change the building entirely.

Not sure how to phrase it? Describe what you want to Claude in chat and ask Claude to write the Claude Code prompt for you. That is a legitimate move, not cheating.

If something breaks

Paste the error into Claude Code. This usually works.
Model won't run — make sure you're in the right folder. Type pwd — it should end in embarcadero-model.
Broke something? Run git checkout . to undo all your changes and start clean. Nothing is permanent.
Still stuck? Ask Claude in chat — describe what you were trying to do and paste what happened.

What's next

Try a project. Build your own personal site with Claude and Claude Code. Mine is one.
Keep at it. You won't solve a big problem on day one. This is a skill and it takes reps.
Be careful with data. Think before you point Claude Code at work files or anything confidential.