How Reducto is building SOTA chart extractions

Reducto's Applied Research team has developed a novel approach to chart extraction, aiming to transform unstructured document data into structured text for enhanced usability in AI applications. This process involves extracting numerical data from charts and graphs, which are common in various industries and document types. The challenge lies in the inherent complexity of charts, which often lack standard formats and contain dense information. Reducto's method involves training specialized lightweight models to interpret different chart components, achieving high-resolution data extraction while maintaining accuracy. This system adapts to specific client needs, such as handling charts without Y-axes, by outputting normalized ratios. Compared to existing solutions, Reducto's approach offers improved accuracy and adaptability, although challenges remain in ensuring 100% accuracy in complex chart scenarios. The company is committed to continuous improvement and invites collaboration with enterprises facing document ingestion challenges.