Ask HN: What is your production ML stack like? (2021)

[AhtiK]

I'm curious about your ML stack that is also used in production. What has failed, what has given joy?

Have you managed to set up a reliable "MLOps" environment with a small(!) team? What are the ingredients?

To what extent do you monitor your model inference performance? Is there an automated KPI tracking in place to make sure the new model architecture or a new set of weights perform as expected?

How much of your deployment has moved to an "ML Cloud"? Whether it's an AWS, GCP or Azure ML-specific services. Which are the ingredients?

[juliensalinas]

Here's the ML stack I have been using for my last project:

- Doing NLP with spaCy (https://spacy.io/) as I consider it to be the most production ready framework for NLP

- Annotating datasets with Prodigy (https://prodi.gy/), a paid tool made by the spaCy team

- Deploying the trained spaCy models onto NLP Cloud (https://nlpcloud.io), a service I helped creating

- Use the models through the NLP Cloud API in production and enrich my Django application out of it