Perhaps the AI trust startup has raised $9 million in seed funding from Andreessen Horowitz to tackle one of the most persistent problems in large-scale language models: hallucinations and factual errors that evade detection before reaching end users.
Founded by Peter Elias, the company aims for 99.99% accuracy, the kind that is common in deterministic software systems but rarely achieved in AI. Its first product is a data science tool that generates answers from complex datasets, each with citations and a full audit trail. The core innovation is what Elias describes as a validator harness. The initial output of the LLM is checked against a deterministic system that rejects results inconsistent with the underlying dataset, and the model is trained in conjunction with its validator to simultaneously optimize speed and accuracy.
This approach has significant commercial advantages. This harness reduces ambiguity with great precision, allowing the system to model and run significantly smaller AI models than Frontier’s equivalents, especially four feature classes below the leading product, enabling deployment on local hardware rather than data center infrastructure, and significantly reducing token costs.
Elias argued that major AI labs have little incentive to solve this level of illusion because their profits depend on the number of modifications and retries their models require. Perhaps the company’s architecture is a reversal of that logic, and Elias said the same engine could extend beyond data science to accounting, medical services and other areas where accuracy is important.
