Cognitiv+

Working with Digital Catapult helped Cognitiv+ to slash its software deployment time by 40% by giving it access to a broad range of expertise. Working together they also explored the potential of semi-supervised learning, which is a leading edge development in the NLP branch of machine learning.

Cognitiv+ needed to conduct in-depth research and development

Companies typically have hundreds of contractual relationships with vendors and clients. Cognitiv+ wanted to expand how artificial intelligence (AI) could be used to accelerate the review of documents, such as contracts that manage these relationships. This would enable companies to rapidly understand and analyse their obligations and risks and make it much easier to retain knowledge about its contracts even when key personnel leave.

Specifically, it wanted to use natural language processing (NLP), an area of computer science that teaches computers to extract knowledge from large archives of text. However, this is a leading edge branch of AI so Cognitiv+ needed to conduct extensive research and development (R&D) to enable it to create a solution.

It had a huge data set of thousands contracts to analyse. To make the best of this dataset, it wanted to explore techniques that are less reliant on a time-consuming labelling phase before the algorithms and model training.

In addition, Cognitiv+’s small team of four did not yet include any DevOps skills which would be needed to make its solution easily deployable on the various configurations clients would need.

Developing an AI system presents numerous challenges beyond a conventional software development project. Code is replaced by inherently large and indivisible models which stretch conventional approaches to deployment and testing. Best practises are still emerging. As a startup with limited funds and a small team, getting access to the range of expertise it needed was a significant challenge that was slowing down product development and time to market for Cognitiv+.

Digital Catapult supported Cognitive+ through collaborative R&D and the provision of data science and DevOps expertise

Digital Catapult offered Cognitiv+ the opportunity to conduct the R&D collaboratively. An activity that was critical to the successful development of the product because it helped Cognitiv+ understand how to use semi-supervised learning to extract knowledge from its huge datasets. Collaborating with Digital Catapult helped to de-risk the R&D process and removed a significant barrier to market entry.

Digital Catapult also provided two data scientists and DevOps support to work with the Cognitiv+ team over a 12 month period.

After a few months of exploratory work, the joint team focused on two critical components, getting the technology stack and processes production-ready and investigating semi-supervised rather than supervised learning approaches.

The engineering team reviewed the Cognitiv+ technology stack and processes to identify areas where they could be streamlined and made production-ready. Like many startups transitioning from R&D to product stage, this is a critical step to ensure efficient, robust, and economic deployment. Cognitiv+ was better able to quantify performance and benefited from the transfer of software best practice skills to the Cognitiv+ core technical team.

Our data science team investigated Cognitiv+’ large corpus of unlabelled data, and highlighted interesting use-cases which arose as a result of the use of semi-supervised rather than supervised learning approaches.

Working with Digital Catapult enabled Cognitiv+ to reduce its software deployment time by 40% and incorporate semi-supervised machine learning techniques into its solution

After a twelve months collaboration, Cognitiv+ launched its solution as a service to market targeting banking, insurance and legal market segments.

This collaboration helped it to hone its value proposition and refine which market segments it wanted to target. It has been able to expand its headcount to 12 full-time employees with a target of reaching 20 in the next year.

The solution is already being used by eight clients and a pilot with Abu Dhabi Global Markets, which is exploring how it could help manage regulatory change, is in progress.

Digital Catapult DevOps and engineering work allowed Cognitiv+ to approach clients with high level security standards that required the solution to be deployed on-premise. As a result of using the tools, it recommended Cognitiv+ has reduced the deployment time by 40%.

Digital Catapult’s data science team provided Cognitiv+ with a general method to address class imbalance in text datasets, applicable to the vast majority of NLP problems in the legal domain such as contract redaction. The performance of the proposed solution was evaluated using employment agreements

Digital Catapult also offered UX support to help Cognitiv+ to develop a clear user interface.