McCarthy Saves 30,000 Hours a Year With a Post-Trained Nvidia Model

Palantir's 21 Oct 2026 release says McCarthy Building Companies uses Foundry, AIP and Ontology. For subcontract drafting, a post-trained Nvidia Nemotron model running on Nvidia compute turned high-level statements of work into Exhibit 1 contract language, delivered better performance than frontier models at significantly lower cost, and saves 30,000 hours a year. Rico Bertucci, vice president of strategic initiatives, said McCarthy built this model on 160 years of McCarthy experience. Separately, McCarthy fine-tuned Nemotron 3 Nano 4B on labeled RFI data and trained on more than 100,000 production RFIs to flag cost and schedule impact. A score chart seen outside the release is not used: the release does not state those percentages or token prices.