The AI Quest

Early 2019 the task was assigned to me to find out how we could incorporate artificial intelligence (AI) in our product line.
The main parts of the product consist of a document management system (DMS) that keeps office documents (Word, Excel, Pdf, email) on SharePoint. A component in the various Microsoft Office product allows the user to assign metadata to the document, and use SharePoint as organization wide file system.
The AI challenge was to be able to suggest both metadata as storage location (which is also a kind of metadata tag) to users for new documents and emails.

The AI challenge was to be able to suggest both metadata as storage location to users for new documents and emails.

Most of our clients are from the legal sector, where 'Matters' are used as the primary classification system for documents.

So the environment to work with is

• A client component hosted in MS Office applications
• SharePoint (both onpremise and online)

I had no previous knowledge of AI, but I soon discovered that most AI research is done using Python, and only a few components and libraries where available within my comfort zone (and that of the Microsoft Platform), which is C#.

Part of that is because AI research requires using the GPU for parallel processing, if you want to spent minutes instead of hours for each AI experiment. This is well incorporated in Python, where Nvidia with the Cuda software is most prominently used.

So while for accessing SharePoint, MSOffice documents and providing web services on IIS one would like to use a .NET language, for the AI environment one has to use Python. This is a nasty gab to bridge. Fortunately Microsoft realized this also only a few years ago, and started to incorporate Python integration possibilities in several ways. Then again, this is all quite new, an not all matured.