Homecastr - AI home price forecasts on a map

Name: Homecastr - AI home price forecasts on a map
Availability: InStock
Author: dhardestylewis

by dhardestylewis·Mar 12, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●MidSlickBold Bet

Claims 8.0% error vs Zillow's 8.4%, but hard to verify without methodology.

Strengths

•FT-Transformer plus Schrödinger Bridge diffusion is genuinely sophisticated ML architecture.
•20+ years of ACS microdata and 8 FRED macro series is substantial training data.

Weaknesses

•0.4% error improvement is marginal and unverifiable without published benchmarks.
•Zillow Zestimate and Redfin Estimate already dominate with far more proprietary data.

Post Description

I built homecastr.com/app because you can't easily see where home prices are headed, not even on Zillow.

I used an FT-Transformer & Schrödinger Bridge diffusion model, trained on 20+ years of ACS microdata & 8 FRED macro series. This approach beats Zillow's baseline on 1-year held-out error, achieving 8.0% vs Zillow's 8.4%.

Happy to answer questions about the architecture or anything else!