Ask HN: Language Agnostic ML

In 2011 I worked for a startup where we approached ML as a compilation process. The learning phase generated a sequence of instructions with coefficients e.g. * a regression step would be compiled into a scalar product with the regression coefficients * a clustering step would be compiled into a nearest neighbour search and a list of cluster vectors etc.

With this kind of approach, it seems “easy” to have a “backend” building the model in a certain language and then have “frontends” doing predictions in possibly other languages. I was quite convinced that this would become the way people build and deploy models.

After a few years outside of Data Science, I'm looking to build models with Python and provide APIs in other programming languages (e.g., Golang) that serve the predictions based on those models. It seems that 8 years later no progress has been made on having more language agnostic approaches to Machine Learning.

Am I completely missing something here?