Valuation Models
Automated valuation systems benchmarked against portal error, segment by segment, with the MAPE figures to prove it.
01 / Division
Applied artificial intelligence for organizations that have to defend their answers. We build the models, valuation systems, and decision tools that turn first-party data into conclusions that hold up under scrutiny.
Capabilities
Automated valuation systems benchmarked against portal error, segment by segment, with the MAPE figures to prove it.
Internal tools that put model output in front of operators in language they act on — not dashboards nobody opens.
Pipelines on our own NVIDIA DGX Spark, not a public cloud, that turn your own records — TEA data, market feeds, transaction history — into a defensible asset.
The controls, documentation, and review process that keep an AI program credible to regulators and clients alike.
Signal
Median absolute error of 7.77% across 19,486 Austin-area sales that closed March–September 2026 and were never seen in training — a forward holdout, not a random split. Single-family homes 7.5%, townhouses 7.2%, condominiums 10.7%. By price band, single-family: 9.9% below $250k, 6.1% to $400k, 7.2% to $600k, 8.4% to $1M, 13.1% above $1M. Stated 95% prediction intervals contain the sale price 89.9% of the time, measured on the same holdout. Published segment by segment because a blended average hides where a model is weak.
Stack
ALEX Intelligence has fully migrated off Google Cloud. Data, models and inference now live on an on-premises NVIDIA DGX Spark — 64 containerised services, no third-party AI in the production path.