Deals found by the latest scan, each priced as a flip and as a rental
The problem
I was learning wholesale real estate from a long video course full of rules of thumb. I wanted those rules as a repeatable, checkable process instead of a feeling.
Claude looks at every photo
Every listing photo is tiled onto contact sheets, and Claude has to open every sheet and return structured findings: condition, likely repairs priced against a local cost table, deal-killers, and what the photos don't show. The code then checks that the number of photos Claude says it reviewed matches the number it was sent before trusting the answer.
One engine, two answers
The same engine prices every house as a flip and as a rental, so the single-listing analyzer and the market scan can never disagree. Comparable sales are filtered by size and outliers, and missing data is labeled thin instead of guessed.
A real scan
One scan matched 349 listings, looked up 120, reviewed every photo of 65, ruled out 35 from what the photos showed, and left 4 worth a phone call.
More screens
The written rules behind the market scan