Winder Research has built a state of the art natural language processing (NLP) platform for a large oil and gas enterprise. This work leveraged a range of cloud-native technologies and sophisticated deep learning-based (DL) machine learning (ML) techniques to deliver a range of applications. Key successes are:
- New NLP workflows developed in hours, not weeks.
- Hugely scalable, from zero to minimise cost to tens of thousands of concurrent connections.
- Enforced corporate governance and unification, without burdening the developer.
Our client, a large enterprise in the oil and gas industry, has thousands of Data Scientists working on a range of problems. This company operates throughout the world and is investing heavily in artificial intelligence (AI) to improve the entire organisation.
Many of these problems require NLP techniques to work with textual data. But individual projects were spending huge amounts of time redeveloping the same tools in tens of different ways. This led to problems in production because operators were trying to maintain similar products that were presented in a multitude of ways. Furthermore, using our platform, new applications were made possible using standard ML patterns. In summary, the company were facing the following challenges:
- People were spending weeks moving simple POC applications into production
- Data scientists were wasting time rebuilding NLP tooling
- New NLP applications were hard to build because they had to be built from scratch
- Some types of NLP applications were impossible to build because technologies weren’t possible
Our solution was based upon the flexible Kubernetes orchestrator, enhanced with the serverless capabilities of KNative. We then built and published a catalogue of NLP services that could be combined to build NLP capabilities. On top of this we implemented a simple, graphical user interface to “wire” services together and built a range of new applications.
The services ranged from standard NLP techniques like entity and named-entity recognition, vectorisation, text searching and document storage. But the most sophisticated examples used DL to build complex textual models that were capable of domain specific translation and explanation, question and answering, enhanced search, web scraping, document retrieval, document parsing and reporting, optical character recognition and many more.
Our solution is in constant production use by the client. Other applications often find there is a service that does precisely what they need, so their development time has shrunk to zero. When new NLP use cases are found, it often only takes a few hours to wire together services, add a new service and release it into production with full corporate IT and security approval. The solutions we developed during this work are also being used throughout the company. In summary:
- Other developers and data scientists can integrate NLP functionality in seconds
- New NLP workflows can be encoded in hours
- Sophisticated DL-model-based applications are enhancing productivity
Our client saw order of magnitude improvements to data scientist and developer efficiency. And internal customers praised the quickly developed sophisticated applications. This led to Winder Research being asked to execute a similar strategy for other DS platform components, further enhancing the clients AI capabilities.
We’ve proven the effectiveness of bespoke internal data science platforms and DL-based NLP techniques are game changing. If you want to improve the efficiency of your data scientists, need a new scalable platform or want to leverage state of the art text models, get in touch.
You might also benefit from our free data science maturity assessment, which helps you prioritise improvements to your business’ data science workflows.