← All client work/Real-estate · Investment

At scaleReal-estate · Investment

Multifamily underwriting software case studyRaw listings became cap rates automatically, with incomplete deals held back and explained.

An investment firm needed to evaluate multifamily properties at scale without slow, error-prone manual underwriting.

ClientREK Partners · Andreas Ortiz

ProjectMultifamily Underwriting & Cap-Rate Platform

My roleSenior developer

StatusAt scale

Discuss a similar build

Property dashboard showing listing cards with cap-rate badges, price, size, unit count and year built, and a "10 missing cap rate" indicator.

The real product, shown with demo data.

~90%Of listings skipped on each run by change detection — only new or changed ones are re-fetched

NOIAnd cap rate from a region-specific expense model

9 linesItemised operating expenses behind every NOI, plus vacancy and a capital reserve

< $40/moTotal running cost for the whole platform: hosting, database, scraping and AI

In short

A real-estate investment firm needed to evaluate multifamily properties at scale. I built a platform that collects property data on a schedule, enriches it with market rents and public records, and computes NOI and cap rate from a regional expense model — and gives no number at all when the inputs are incomplete.

01 · The situation

Underwriting by hand was slow and error-prone.

A real-estate investment firm needed to evaluate multifamily properties at scale. Manually pulling listings, rents, and public records, then running the underwriting math, was slow and error-prone.

02 · The challenge

A wrong cap rate means a bad investment.

The data lives behind anti-bot defenses and in inconsistent formats, and the financial model has to be exactly right. So it needed reliable data collection and correct, region-aware financial calculation.

03 · What I built

From raw listing to investment metric, automatically.

A platform that automatically collects property data on a schedule, enriches each property with market rents and public records, classifies and analyzes it with AI, and computes investment metrics — surfaced in a dashboard with filtering and sorting, where every figure can be opened to see the calculation behind it.

Side panel for a 28-unit apartment building priced at $3,990,000 with a 6.89% cap rate and tabs for overview, units, cap rate, calculations, public records and images. — A property's detail panel: price, cap rate, units and size, with tabs for units, calculations and public records.

Calculations tab listing annual gross potential rent, vacancy loss, nine itemised operating expenses, capital reserve, net operating income and a 6.89% cap rate. — The full calculation behind the number: revenue, vacancy, nine expense lines, reserve, NOI and cap rate.

Stats page with cards for active properties, rent, cap-rate, parcel and image coverage, and a chart breaking down the reasons a cap rate is missing. — Coverage at a glance, and a breakdown of why any listing has no cap rate yet.

Settings page with versioned, city-specific cap-rate parameters such as vacancy rate, management fee, tax ratio and insurance base. — Expense assumptions are versioned settings per region, not constants in code.

Formulas page explaining gross potential rent, vacancy loss, effective gross income and operating expenses, each with its formula. — Every formula is documented in the product, with the live values it currently uses.

Screens from the real product, captured with demo data. No client or customer data is shown.

04 · How it works

A scheduled underwriting pipeline

01CollectListings behind anti-bot defensesChange detectionRetries
02EnrichMarket rents + public records
03AIClassifyProperty type from images, details from textVisionGPT
04Cap-rate engineNOI from a region-specific expense model
05Completeness checkNo price, units or rents: no cap rate
06DashboardFilter, sort, review

On a schedule the platform collects new and changed listings, enriches them, classifies them with AI and computes the investment math — and leaves the cap rate empty, with a reason, when an input is missing.

05 · The hard engineering

Reliable data, correct math.

The cap-rate engine, one property at a timeIllustrative figures

Gross scheduled rent24 units × $1,350 × 12$388,800

− Vacancy (5%)−$19,440

− Operating expensesRegion model: Texas−$155,131

= Net operating income$214,229

÷ Asking price$3,150,000

Cap rate6.80%

Completeness check

✓ Price found

✓ Unit mix found

✓ Rents found for the unit mix

⚑ No price: no cap rate

⚑ No rents yet: no cap rate

A listing that fails a check gets no cap rate, and a stated reason.

Resilient scraping

Survives anti-bot systems and changing page structures, with retries and monitoring that surface collection failures.

AI used precisely

Image-based property-type classification, plus extraction of unit counts, square footage and key details from listing text.

Cap-rate engine

Per-unit market rents fed into an engine that computes NOI and cap rate from a detailed, region-specific expense model.

No guessed numbers

A listing with no price, no unit mix or no rents gets no cap rate, and the stats page says which reason applies — instead of a figure that looks right and is not.

Change detection

Each run re-fetches only listings that are new or whose price or status changed. That skips about 90% of them, which is why the schedule moved from daily to weekly.

06 · Key decisions

The calls that shaped the build

Why is the financial model region-aware?

Operating expenses differ by region. A single national assumption would produce cap rates that look right but lead to bad investment decisions.

Where exactly is AI used?

Only where it adds something rules can’t: classifying property type from images and pulling unit counts, square footage and key details from listing text.

Why leave a cap rate empty?

A wrong number is worse than no number. When a listing has no price, no unit mix or no rents, the platform shows no cap rate and says which input is missing, instead of silently skewing the results.

07 · The outcome