Beyond Work Unveils AI Agent for Document Intelligence

The AI agent streamlines document processing by cutting manual labour and operational costs.

Beyond Work, an enterprise AI company, announced the results of Matrix, a memory-augmented AI framework for automating business document processing.

Developed in collaboration with researchers from Penn State University, Oregon State University, and Kuehne+Nagel, one of the world’s largest logistics providers, Matrix addresses the complex, time-intensive task of extracting transport references from Universal Business Language (UBL) invoices.

By harnessing an iterative, memory-centric learning strategy, Matrix achieves a 30.3% improvement over chain-of-thought prompting, outperforms a standard Large Language Model agent by 35.2%, and surpasses Reflexion by 27.28%-establishing its state-of-the-art capabilities in AI reflection.

“Matrix redefines what’s possible for enterprise automation by dramatically improving accuracy while reducing operational costs,” said Malte Højmark Bertelsen, co-author and cofounder of Beyond Work.

Matrix’s success is the result of an international team of experts, including Jiale Liu, Yifan Zeng, Malte Højmark-Bertelsen, Marie Normann Gadeberg, Huazheng Wang, and Qingyun Wu, an Assistant Professor at Penn State University recognised for her contributions to Automated Machine Learning (AutoML) and Large Language Models (LLMs).

Her track record includes high-impact open-source projects, such as AutoGen, that enable complex multi-agent collaborations – foundational principles driving Matrix’s memory-augmented approach.