// jays.website / architecture / rental

Rental Intelligence, Under the Hood

Every multifamily building in Massachusetts — 339,856 of them — assembled into one spatial database, scored 0–100 against real rent benchmarks, and connected to an ownership graph of 30,000+ landlords. Not a listings site: a decision engine that turns a state's worth of parcels into a shortlist.

The Pipeline
MassGIS parcels→ assessor + HUD FMR join→ PostgreSQL / PostGIS→ composite deal score→ ownership graph→ static export→ county rankings + map

A daily refresh re-scores the dataset and auto-commits the exported data — the site updates itself.

Data Assembly

A state's worth of buildings in one schema

Parcels · Assessments · Benchmarks

Sources

  • MassGIS statewide parcel data — the authoritative geometry and use-code layer for all 14 counties
  • Town assessor records: unit counts, year built, assessed value per building
  • HUD Fair Market Rents by county and bedroom count as the rent benchmark

Storage

  • PostgreSQL with PostGIS — spatial joins, county rollups, and geometry queries in one engine
  • Analytics tables keyed per building so scoring, comps, and ownership all resolve to the same record
  • Persistent data lives on dedicated storage, backed up separately from the code

Hygiene

  • Every score carries a data-quality confidence component — thin assessor data can't masquerade as a strong signal
  • Coordinate-bounds filtering keeps bad geocodes out of the map layer
  • Schema keys are enforced consistently across services after early mismatches proved how silently joins can rot
Scoring Engine

From 339,856 buildings to a shortlist

Yield · Distress · Ownership · Confidence

The composite score

  • A 0–100 deal score per building: estimated yield against the HUD benchmark, distress indicators, ownership signals, and data confidence
  • A rent-estimation pipeline seeds per-building rent figures where no listing data exists
  • Two-layer ranking separates "good building" from "good opportunity right now"

Ownership graph

  • Owner-name resolution collapses 339k parcel records into 30,000+ distinct owners
  • Portfolio holders, absentee owners, and institutional-scale landlords are flagged — who owns a building is itself a signal
  • Comparable-sales service benchmarks each candidate against nearby transactions

Human in the loop

  • The engine produces a shortlist, not a verdict — final judgment stays human
  • A review-labeling loop feeds real accept/reject decisions back toward the ranking weights
  • Deal-flow candidates route through a chat-based approval gate before anything is acted on
339,856
Buildings
177k
Scored
30k+
Owners mapped
14
Counties
Delivery

Heavy database, light site

Static Export · Daily Refresh

Outputs

  • County opportunity rankings, top-scored buildings, and an interactive statewide map filterable to street level
  • The site reads pre-exported static data — the database never sits in the request path
  • Free to explore; a Pro tier (alerts, CSV export, cap-rate calculator) is in the works

Automation

  • A daily scheduled refresh re-scores the dataset and commits the updated export automatically
  • Exports are versioned in git, so every day's rankings are reproducible
  • The same pipeline pattern — heavy compute at home, static delivery at the edge — as the trading system

Explore the rankings live, or step back to the full systems tour.

Rental Engine → ← All architecture