Coffee Sessions #157

MLOps Build or Buy, Startup vs. Enterprise?

There are a bunch of challenges with building useful machine learning at a B2B software company like Slack, but we've built some cool use cases over the years, particularly around recommendations. One of the key challenges is how to train powerful models while being prudent stewards of our clients' essential business data, and how to do so while respecting the increasingly complex landscape of international data regulation.

Take-aways

Experience when learning MLOps: - all about third-party tools and platforms for hyperparameter tuning and model serving, e.g. DVC, MLFlow, Weight & Biases, TensorFlow Serving, Kubeflow, SageMaker, feast, HuggingFace - there's mention of monitoring yet no specific tools in the MLOps land that target monitoring as far as I recall The impression during the time is that there are so many tools we can utilize, I don't know which, but there must be something. Reality working in Enterprise: - minimal hyperparameter tuning - so many things/dashboards to monitor! - constraints in infrastructure to integrate with third-party tools - the pro: existing and mature infrastructure that we can utilize, e.g. data warehouse, monitoring (logstash + prometheus + honeycomb) So the team ended up building or migrating to its own infrastructure most of the time.

In this episode

Katrina Ni

Katrina Ni

Senior Machine Learning Engineer, Slack

Katrina Ni is a Machine Learning Engineer in Slack ML Services Team where they build ML platforms and integrate ML, e.g. Recommend API, Spam Detection, across product functionalities. Prior to Slack, she is a Software Engineer in Tableau Explain Data Team where they build tools that utilize statistical models and propose possible explanations to help users inspect, uncover, and dig deeper into the viz.

Twitter

LinkedIn

Aaron Maurer

Aaron Maurer

Senior Engineering Manager, Slack

Katrina Ni is a Machine Learning Engineer in Slack ML Services Team where they build ML platforms and integrate ML, e.g. Recommend API, Spam Detection, across product functionalities. Prior to Slack, she is a Software Engineer in Tableau Explain Data Team where they build tools that utilize statistical models and propose possible explanations to help users inspect, uncover, and dig deeper into the viz.

LinkedIn

Demetrios Brinkmann

Demetrios Brinkmann

Host

Demetrios is one of the main organizers of the MLOps community and currently resides in a small town outside Frankfurt, Germany. He is an avid traveller who taught English as a second language to see the world and learn about new cultures. Demetrios fell into the Machine Learning Operations world, and since, has interviewed the leading names around MLOps, Data Science, and ML. Since diving into the nitty-gritty of Machine Learning Operations he felt a strong calling to explore the ethical issues surrounding ML. When he is not conducting interviews you can find him making stone stacking with his daughter in the woods or playing the ukulele by the campfire.

Jake Noble

Jake Noble

Host

Jake is a tech lead on the Product Engineering team at Tecton. Before Tecton, Jake worked at YouTube on Homepage recommendations and client infrastructure for six years.