What happened
Probably, led by founder Peter Elias, secured $9 million in seed funding from Andreessen Horowitz to develop AI systems achieving 99.99% accuracy by preventing hallucinations. Its initial product, a data science tool, employs a "data science mech suit" — a deterministic validator system — to verify LLM outputs against datasets, ensuring results include citations and audit trails. This approach enables the use of smaller models, "four classes weaker" than frontier models, capable of running on local hardware.
Why it matters
Access to highly accurate, verifiable AI outputs will reduce operational risk for procurement teams and security architects in precision-sensitive sectors like accounting and healthcare. The mechanism of deterministic validation allows for deployment on local hardware with smaller models, cutting token costs and enabling on-premises inference for sensitive data. This contrasts with the increasing compute costs for frontier models, offering a cost-effective path to AI reliability, a critical need following instances like xAI Grok inducing user delusions.
