What's Happening?
Docusign has developed a new AI model to improve the efficiency of processing over one million documents daily. The company has implemented a teacher-student model approach, where a larger model generates high-quality labels, and a smaller model is trained
on these pseudo-labels. This method allows Docusign to maintain accuracy while reducing costs and latency. The AI team at Docusign has ensured that data privacy is maintained by anonymizing data before it is used for training. The company has also developed a smart sampling pipeline to ensure diverse and representative datasets for model training.
Why It's Important?
The implementation of this AI model is significant as it addresses the challenge of processing large volumes of documents efficiently without compromising on accuracy. By reducing costs and latency, Docusign can offer more competitive services to its clients. This development also highlights the importance of data privacy and the need for innovative solutions in handling large datasets. The approach taken by Docusign could set a precedent for other companies in the industry, emphasizing the balance between efficiency and data protection.
What's Next?
Docusign plans to continue iterating on its AI models, with a focus on reinforcement fine-tuning using newer reasoning models. The company is also looking to extend its platform to allow enterprises to define custom extractions, which could further enhance the utility and flexibility of its services. This ongoing development suggests that Docusign is committed to staying at the forefront of AI technology in document processing.
