PermitNerd

PermitNerd — A construction permit lifecycle and prediction dataset generated by an expediting workflow SaaS. Three distinct assets. First, estimator-versus-outcome pairs: the Permit Estimator predicts required permits, review departments, fees, timelines and risks for a described project, and the live city sync then reveals which permits were actually filed, issued, delayed or rejected, so every estimate has a matched ground-truth outcome. Second, blueprint document pairs: customer drawing uploads with the automatically extracted square footage, room counts and project details, plus the esti…

PermitNerd offers (Alternative, Supply Chain, Reference) — Modest and concentrated. Saved-permit ceilings of 100, 500 and 1,500 per account across tiers imply a tracked-permit universe plausibly in the tens of thousands to low hundreds of thousands of records given the inferred account base. Blueprint analyses are rate-limited to 5 or 10 per month on paid tiers only and estimator reports to 2, 10 or 20 per month, so the drawing corpus and the prediction-versus-outcome pair set are each likely in the low thousands cumulatively rather than at scale — approximately 2,000-10,000 blueprint analyses and a similar order of estimator reports since launch, both bounded by per-account monthly caps..

Document-understanding and vision-training data for architectural drawing comprehension — spatial reasoning over plans with attribute ground truth, which is scarce and expensive to label by hand. Tabular and structured-output model training for regulatory requirement prediction from a free-text project description. Fee and timeline forecasting models for construction estimating. State-transition and survival models for municipal approval latency, panelized across 10 jurisdictions with very different distributions. Domain fine-tuning and preference or rejection-sampling data for construction-regulatory language models from the estimator reports and the notice-decoding dialogue. A genuinely hard evaluation benchmark: held-out projects where the permit set was predicted before the outcome was known, which is a clean prequential test set that cannot be reconstructed after the fact.

The data is with 3 years of history.

Coverage spans US; Construction & Engineering, Real Estate Services; alternative, supply_chain, reference; equities.