PROLABS.AI

In the last decades, as massive amounts of relevant, timely and high quality data finally have become available for most modern enterprises, and as computer processing power has soared, it finally became possible to use sophisticated machine learning models to automate various business processes. Even more so, modern enterprises are contemplating automating some mission-critical tasks that have historically been very labour intensive. And they are doing it with machine learning, a technology tha... t has matured over the years and now there are many successful cases when complex models have dramatically increased business efficiency. However, modern machine learning models are quite complex, may consist of sophisticated pipelines of algorithms, consume a lot of resources and may appear to the business as complete black boxes. The former problem is particularly challenging, since it raises concerns about relying on these models for decision making without really understanding how decisions are made. Overall, mission critical applications require much more than good models to operate. They need an environment that is both reliable and will shield the application from external attacks, fraud, faults in the upstream and external systems and dramatic sudden changes in operating conditions. Furthermore, in mission critical applications there should be multiple models available on standby, which include weaker, but more robust models. All the feature extractions pipelines need to be carefully managed and versioned, such that models can be easily compared and model improvement is a straightforward process. The data that the models operate on must be complete and of highest possible quality. Model features and outputs must be intelligently monitored and automatic alerts need to be set up in order to quickly detected changes in market conditions, fraud waves or technical malfunctions in the upstream systems. Intelligent monitoring and an ability to fall back to simpler and more robust models can mitigate the issue with using complex black-box models such as neural networks. Prolabs has developed its machine learning platform Fabrique to address these issues, including shielding the application from various security risks. Prolabs is also building a number of vertical solutions using Fabrique, currently focusing on Fintech and Predictive Maintenance applications.More