Built by the Community
Voting function, podcast, release, hiring!
Adam here, with my first time giving you all an update. Here’s the next version of the newsletter where we talk about daily going-on’s of building an anti-stealth startup. We keep it short and snappy. Let’s go!
1. Don’t boo. Vote! 🗳️
ZenML is meant to be built by the community. As an open-source and anti-stealth startup, we are not building a product for any particular company. We are doing this for data scientists with daily MLOps challenges. This is why we established a responsive feedback loop with a voting function on GitHub. Every weekly sprint the ZenML team is filling the sprint backlog with the requests of the community. So we make sure not to lose track of what we really need to build and are not developing something for the drawer. We have limited resources and need to allocate them most effectively. So, build ZenML with us and make your voice heard:
We have several options for you to influence the roadmap:
Vote on your most wanted feature on the Discussion board.
Create a Feature Issue in the GitHub board.
Start a thread in the Slack channel.
2. Introducing Pipeline Conversations 🎤
Alex initiated our own MLOps podcast called “Pipeline Conversations”. The first episode was about the podcast itself, ZenML, and why we are doing this with Hamza and me. We just recorded the second version with a “real” guest and publish it next week - stay tuned and subscribe here.
We’re filling up the funnel of the next guests and are really looking forward to the current line-up. We want this podcast to be more technical and real, grounded with conversations with real-world data scientists and engineers struggling with MLOps challenges. If you have any interesting candidates, please send them to us via firstname.lastname@example.org.
3. Release ZenML 0.5.3 and how LinkedIn is so unpredictable 😲:
Like every two weeks, we also have a new release this time. Let’s not talk about the actual release as you can find all the details here. Let me share something else:
I was casually posting about it on LinkedIn and got almost 9k views and 100 reactions within the first 24h. It’s fairly good for a “simple” post without real content. Maybe it was the GIF.
It really goes to show—with online content it’s about being consistent: You never know what strikes a chord!
4. Keep hiring! 👩👩👧👦
Initially, we thought 10 people would be enough to grow the team for the start. But now we found so many more angles on how to create value in the MLOps space. Specifically, the idea of our ZenHacks hit a nerve. (ZenHacks are days when the ZenML team is completely detaching from the actual development and using a real example or research paper to battle test ZenML).
So we’re now looking for 2-3 working students to consistently create ZenHacks all day long and productionalize repositories from resources like Papers With Code. It will be fun to show the capabilities (and boundaries) of ZenML and give people access to state-of-the-art research with a reproducible MLOps pipeline.
BTW on hiring: Next week we have two full-time joiners! Stay tuned for their introduction! We’re really stoked about building up the core team.
5. Bytedance and GPU coil whines
Speaking of ZenHacks, we had a lot of fun tackling two papers in the last two weeks, CycleGan (image-to-image translation) and the new ByteTrack model from Bytedance (multi-object tracking). We will follow up with a more thorough blog post about both, but Bytedance in particular gave us a lot of pain and was a real-world example of how hard it is to get state of the art models in any semblance of real-world production systems. It got so bad at some point, that Baris’s Nvidia GeForce RTX 3090 started literally (coil) whining from the strain!
P.S. If you think his machine setup is amazing, do feel free to forward this newsletter to your data science friends!
Alright, that’s a wrap for the last two weeks, so let’s end it here this time.
お疲れ 様 です,