Sansan Inc. announced its selection for the 'Post-5G Information and Communication System Infrastructure Enhancement R&D Project (GENIAC)' promoted by METI and NEDO. Following last year's phase 3, Sansan will develop a unique visual language model capable of accurately responding to complex instructions in Japanese business documents like invoices and contracts. Sansan has refined its technology to digitize analog business documents by combining machine learning and human quality assurance, creating high-quality primary information crucial for AI performance. Their core technology, the multimodal generative AI 'Viola', extracts items from document images. In GENIAC phase 3, they developed 'Cello', identifying information coordinates to suppress hallucinations. In phase 4, Sansan proposed developing a visual language model with advanced instruction-following capabilities. The model will be integrated into 'Bill One' and 'Contract One' to drive back-office automation and address the shrinking workforce. Sansan also plans to develop and publish a benchmark for similar tasks with Nara Institute of Science and Technology.
FACT BOX
- Source: PR TIMES
- Category: business