Most of the recent feature set added to the CloudSinc platform have been either directly or in support of machine learning (ML). The more I apply it, the greater the scope I see for it to be used throughout apps and indeed across business processes.

As Andre Karpathy discussed in Software 2.0 5 years ago, the core, difficult algorithms (image/sound/game-play) are increasingly ML based. My view is that while this is impressive, there’s a huge (and certainly the majority of accessible usage) opportunity to apply statistics via ML tools to everyday business data. The fact this data is (mostly) structured, means fast training and quick results.

Initially I was keen to add UX and config automation to default actions based on previous behaviour — this is a fairly common usage of ML. For example, if the last time a break (ie a data exception) occurred and you/another user actioned it in a specific way (sending an email, assigning a ticket to a team), then making the UI adjust to make it quicker next time (highlights, quick-action buttons), is a straight-forward and useful approach.

Next up was root cause analysis — at least to the extent possible in the data available. This is essentially pattern finding and presenting the results in a form digestible to a user. The early proof of concept work can be seen on this demo page.

This work spawned much thinking about presentation of ML results. While Sankey and flow charts are great for user understanding and interpretation adding source code output was a step change (see below for a list of projects doing similar).

The great advantage of code output is that the model is realised in a manner that’s transparent (for those who read that particular language), verifiable (no hidden paths, though that’s different from 100% correct) and actionable. This route also allows adoption via integration more easily with 3rd party tools.

Currently we can generate:

  • Sankey charts
  • Excel formulas (for user testing/understanding/lightweight usage)
  • Python (standard data processing language)
  • SQL (for those using dbt models or a datawarehouse such as Snowflake)
  • Java (general purpose language common in larger firms)

We will add more as customers request (initial requests around SAP and Azure data services, for example).

I’ll finish here and continue thoughts in another post.

Other ML code generation tools

A collection of tools and companies delivering or chasing AI software production is shown in this thread.