We built a document AI that reads but doesn't write (www.nutrient.io)

Nutrient has announced the launch of its new document decision model, which processes documents by directly answering structured questions with specific choices and confidence scores, without generating text responses like traditional large language models (LLMs). This model can discern document types (e.g., invoices vs. contracts) and evaluate the presence of specific data, making it particularly suited for tasks such as verifying extracted data, classifying documents, and splitting merged PDFs. With response times as quick as 200 milliseconds, the decision model significantly reduces both latency and operational costs compared to LLMs, which can take longer and incur higher costs due to text generation. The significance of this model lies in its ability to streamline document-based workflows that require rapid, accurate decision-making without needing verbose explanations. By integrating both text and image inputs, Nutrient's model enhances its recognition capabilities for visually dependent tasks, such as distinguishing between different types of charts in a document. Furthermore, it allows for intricate tasks like evaluating confidence levels in answers, providing a more nuanced understanding of its outputs. This approach opens up new avenues for optimizing AI applications in document processing, reinforcing the potential for decision models to coexist and complement LLMs in AI ecosystems.