2021 Retro

Neoway has developed an in-house Feature Store to make data scientists’ work easier during the model development either to discover datasets available for generating features or to create new feature datasets to be used on the models. They have been developing tools for improving their feature pipelines since when the term ‘feature store’ was not popular and that learning journey helped them in this new development.

The company has a data platform that provides services to collect data from external sources, apply treatment functions to produce curated business entities, and make them available for consumer applications. The main data platform components are a Schema Registry for improving data governance, a Data Lake for offline data storage, a Kafka for data streaming, and data capabilities such as PostgreSQL as the relational database, BigQuery as the data warehouse, ElasticSearch as the search engine and MongoDB as the online and fast data storage for APIs and applications.

Neoway's Feature Store is fully integrated with the company’s data platform, and this allows the interaction with data capabilities to perform tasks very easily, such as reading files from and writing files to the data lake, registering schemas in the Schema API, producing data to Kafka, inserting records in PostgreSQL, indexing data in the Elastic Search and making them available for customers in company’s SaaS and APIs. Read the full article now to find out more.

I've been meaning to write about the work our friends Mignev and Vesi have been doing at TeachableHub for months. No more procrastinating in the new year!

So what is TeachableHub? It's a platform focused on the deployment side of MLOps with a special emphasis on collaboration. There are so many stakeholders involved in the ML process, the Teachable crew is rethinking how to make it easy for everyone involved.

Mignev told me "We want you to deploy your models a couple of times a day, not a couple times per month." Bold statement seeing as how some of us would be happy to be deploying a couple times per month even.

What I love about how the team attacks MLOps problems is they are very user focused. This is evident by the blog posts they put out, and also they listen to the community podcast, hear what people are having pains around, and build products for that! An example is in one coffee session Vishnu spoke about how its complicated to write a bunch of code to test your model every time you want to deploy it. The Teachable crew heard this and created a deployment scheme and baked in verification tests by default.

"Our goal is for you to not have to document anything from this day forward."

Another bold claim. So what does TeachableHub do to make that a reality? The track and automatically document the following:

host name of the server,
what framework, requirements, dependencies
versions of python
tracks changes
all autogenerate info about the API

how to use it, how to integrate it, info about all the features
how to integrate rest API
what are the results and the errors

This is a tool that's meant for small teams of data scientists looking to collaborate more effectively and they don't have the support from MLE or the DevOps team. The Teachable team has incorporated their 8 years of experience serving over 110 million devices at Cloudstrap to create this tool. As Mignev told me when we spoke "At the end of the day we need to deploy software, not just serve models".

I am looking forward to seeing what this team can create, they came out with version 1.0 towards the end of last year. check them out by clicking the link below.

**This is NOT sponsored and I choose these tools to write about based on projects I hear about and feel deserve a shoutout. Reach out to me if you think your tool would be a good fit.

What a year it’s been! When we we took stock of this community, Demetrios and I were shocked at the scale it had already reached (~2500 members or so). Today, we’re larger, deeper, and more helpful than ever before thanks to the participation of you great machine learning professionals. So let me start the recap of this episode with a personal thank you!

Demetrios and I took some time in this episode to share some of our best lessons, reflections, and reasons to be excited for 2022. Both of us had a lot to say, especially about how we’ve learned a great deal about the data component of the machine learning process. Demetrios provided a super comprehensive overview of all that happened in the MLOps ecosystem. We closed out with some community news: write blogs for us and join our organizing team to make the community a better place!

Give this one a listen to take stock of where we’re at with MLOps in 2021 and where we hope to go next.
- Vishnu

Austin Texas, January 27th & 28th. Data Council. In Real Life.

I've got one ticket to give away to the perosn who replies with the best war story. Current ticket value is at $1299. You still gotta get there and pay for logging but considering the lineup it looks well worth it. The war stories will be shared with the community after so please don't put in any incriminating or identifiable info that could get you fired/sued.

If you wanna just buy the tickets and not share your stories with us, you can get 20% off by using the promo code MLOPSDCAUS22 at the checkout. I think there will be a few community members there so if you go let us know!