Case Study: EstateMatch — Turning PDF Exposés Into AI Investment Verdicts

German real-estate investors were doing the same thing over and over: open a PDF exposé, copy dozens of numbers into a spreadsheet, and run hours of tax and cashflow math before deciding whether a deal was worth pursuing. EstateMatch replaces that entire ritual.

An investor uploads the exposé and Claude AI extracts 50+ fields automatically — purchase price, rents, financing terms, AfA rates, and the federal state. From there a calculation engine applies real German methodology: Grunderwerbsteuer by state, degressive AfA, the §7b and §7i special depreciation rules, bullet loans, and a full ten-year cashflow and IRR projection.

The part I'm most proud of is how the verdict is generated. The deal score (0–100) is computed deterministically from the numbers, and the BUY / HOLD / PASS call is forced from that score — so the recommendation is never a hallucination. Claude only writes the prose rationale around a decision the math already made.

I shipped it end to end as the sole developer: the extraction pipeline, the engine, an admin console, a deal marketplace, shareable links, A4 PDF export, and a fully bilingual EN/DE interface. It saves roughly six hours of manual work per deal.