HHD Real Estate AI Brief

Framework detects five entity classes — columns, beams, walls, braces, openings — without task-specific detector training.

Scale estimated within 0.1% of generator reference on every drawing in the held-out test set.

Recall/precision: 0.922/0.997 for columns, 1.000/1.000 for walls and braces across 50 held-out plans.

AI was unable to search 16 of 200 files — 8% — due to absent title plant data or normalised datasets.

Involuntary lien detection carried an issue fail rate exceeding 36%, the highest-risk gap identified.

DataTrace manages more than 2,000 title plants nationally and self-funded the study.

S3 Ventures led the round; Mischief VC — fund of Plaid CEO Zach Perret — participated alongside three repeat backers.

BoomCRM is built on Boom's existing underwriting engine, targeting agentic leasing and touring workflows.

Boom covers roughly 500,000 units across more than 400 operators, including 25%+ of the top 40 single-family rental (SFR) third-party operators.