Mastering their clouds and scaling their AI infra with Zeet

Mastering their clouds and scaling their AI infra with Zeet
< 1

week to onboard


deployments in first month on Zeet


Clusters managed by Zeet


Clouds managed by Zeet


If you're interested in learning about how Julius uses Zeet to handle their cloud infrastructure, you're in the right place! And if you have cloud infrastructure or Kubernetes workloads you want to simplifying deploying to and operating, learn more about how we can help or try Zeet for free!

Using AI for data science isn’t a new concept. To create AI, data scientists created models that could make sense of massive datasets, and using these findings, they built better and better models that they could leverage to answer more and more complex questions. What we’re seeing now, however, is different—now anyone can use these models and their data to uncover insights.

From small mom and pop shops, to large multi-national corporations, AI is making data science accessible to everyone. “We can shift from spending time doing manual tasks…and move to more cognitive techniques where we can analyze the data because the computer is doing much of the busy and manual work.”

George Forbes
Director, HAF Digital Operations Directorate, U.S. Air Force


No longer do you need a data scientist to run the models. AI knows the models, can learn your data, and can use the models on your data to get you the insights you want. If the US government is on board, you know it’s gone through the red tape, and with tools like this becoming more pervasive, affordable, and accessible, the future of data science is looking rosy.

Enter Julius AI. Julius is an AI-powered data science assistant that can provide first-of-its-kind analysis on structured data sources without human intervention. Julius started on a PaaS, but it soon became clear that they’d need more flexibility and control, leading their small but mighty team to make the switch to Zeet. With Zeet, they can manage thousands of jobs, prototype with ease, and focus on building, all while their user base continues to scale.

Julius AI has a small data science team that wanted to try something that seemingly should have been straight forward—structured data analysis. You’d think, structured data analysis is the easy stuff, right? We’ve had programs and packages that facilitate this, so why is this interesting? Well, it’s interesting because AI is great at unstructured (relative to humans), but when it comes to structured data, this is thus far mostly uncharted territory.

Just a few months ago they started Julius, and after overwhelming interest from the market, what was just a prototype had to become a product (and business) almost overnight.

As mentioned, they started on a PaaS, and this worked, for a few weeks. They found the PaaS to be great in the prototype stage because they could focus on building, not on infrastructure, however as soon as they started actually getting traffic, cost and scaling constraints concluded their PaaS journey.

Their PaaS was the first constraint, and it became problematic quickly: it only allowed 100 database connections. At this point, they had no choice, so they began the laborious process of moving their infrastructure over to GCP. This was not a tractable problem for their small team, and realizing they’d be easily be spending $1000+, a more flexible developer platform started to be a no-brainer. Their founding engineer, Matt Brockman, was familiar with Zeet, having used it at a previous company, so the choice was simple.

“If you’re going to pay an engineer $200/hour, and it takes more than 5 hrs, just use Zeet. It’s basic cost savings and it works.” - Matt Brockman, Founding Engineer

The switch was simple, and the benefits extended beyond letting them manage infrastructure. They could scale effortlessly, and the prototyping and iteration speed they were used to on their PaaS was similar with Zeet. They could perform maintenance, roll out changes to their infra and code without downtime or cold starts, and do all this from one dashboard, no matter how many cloud providers they eventually bring on.

“We spent a whole weekend working on containerized workflows, and while we initially had to learn how to use Blueprints to accomplish this, with the help of the Zeet support team, we got it working quickly…migration is easy.”

“We can focus on the code, not the infrastructure. There is only so much time in a day, and even if a thing takes 30 seconds, context switching is expensive, and having everything in one place just makes things so easy. Zeet is easy to use, and it saves us time so we can focus on…building our product, and with your support offering, we feel like we can keep moving—it’s like our own personal ChatGPT.”

Alex Kuo
Founding Engineer


Caesar isn’t done integrating, but they made sure to emphasize, it’s not because they couldn’t. Migrating is simple, but coming from a PaaS and from GCP at the same time, with all the constraints and complexity, while maintaining security protocols is never straightforward. Second time founders often use Zeet for this reason. As they said, “Migration is easy”.

So how have they been using Zeet? Well, they’ve been using it in pretty much every way you can.

Zeet Jobs allows you to create a “job” from a Project, have that run once, complete whatever task its designed to do, and then self-destruct. Caesar was running up to 500 jobs a day. After some reconfiguring of their stack to maintain state within their models, they are now doing all of this using a Jupyterhub Helm Chart and custom Docker Image.

The kicker was they deployed it all in a fraction of the time by leveraging Zeet’s Helm Chart Blueprint. They then used a different Zeet Service to deploy different Docker Images for each Jupyter server to support each customer. They plan to eventually write custom IaC packages to make this even more simple and scalable, but for now, they are happy with the gains Zeet unlocked.

Beyond serving customers, with Zeet they are better able to harness the power of their cloud, GCP. As they and many others have attested, it’s a huge learning curve to work in a cloud console, and the big advantage of a platform is if and when they expand to other clouds, they won’t have to learn another cloud console, they already know Zeet.

“Starting from scratch pre-revenue, a PaaS makes sense when you can spend $50/mo on infra and focus on building and prototyping, but as soon as you need to scale at all, you move to Zeet. You can still prototype on Zeet, but once you start getting to scaling, reliability, and metrics, you’d use Zeet.”

Matt Brockman
Founding Engineer


Zeet works for Caesar. Whether it be prototyping, managing infrastructure, security, control, or scaling, Zeet allows Caesar to get the most out of their clouds present and future, and focus on building their product, no matter how big they get.

Blueprints Used
Kubernetes Container App

Using the official Kubernetes Application Blueprint, you can deploy your docker apps in just a few clicks.


What does your team’s production readiness practice look like for deploying a new service? In this article, we’ll share some helpful approaches for creating a great production readiness checklist.

Docker Image

Deploy a Docker image from the Docker Hub to any Kubernetes cluster.