Introduction:  

The previous article introduced some of the essential steps in getting a small lightweight eCommerce website up and running in minimal time. We mentioned that the users for this website may need to be recognized, their transactions on the website may need to be remembered and the processing for each order may need to be transparent to the user.  This works well for in-house software development across industries for a variety of applications using off the shelf products, cloud services and development framework and tools. Web applications and services fall in this category. Most software engineering development in industries such as finance, retail, telecommunication and insurance have a significant amount of domain expertise and picking the right set of tools, resources and processes is easy for the business sponsors and their implementors. 

However, when the domain remains the same and we apply new computational capabilities that require significant new knowledge and expertise such as in machine learning, then it is slower to onboard and expect new team members to realize the applications. In such cases, the machine learning toolkit provider may only be able to put out samples and community news and updates. The companies are then best served by white-glove service that not only brings in the expertise but also delivers on the execution. First it reduces the time to implementation because the skills, resources and best practice are available. Second, the challenges are no-longer unknown and have been dealt with earlier in other companies. These together argue for a specialized consultancy services in machine learning development in most verticals.  Even web-application development started out this way in many organizations before having indigenous employees assume all application development  effort. Some organizations may want to have both - the expediency to realize near term goals and the investment to build long term capabilities.

I have a sample application : http://shrink-text.westus2.cloudapp.azure.com:8668/add to illustrate an example. Suppose we want to use this as a sample within a particular domain, then we would need to justify it over say SQL Server text classification capability. Here we need not argue that the above processing does not require text data to make its way into the SQL server for the above service to be used. Instead, we focus on the model tuning and customization we can do for the same algorithm as in SQL Server as well as model enhancement with other algorithms say from R package while allowing to operate on data in transit in both cases. 

#codingexercise

Tests for https://ideone.com/6ms4Vz