ZenML aims to serve as the binding force that unites various open-source AI tools. This open-source framework enables the creation of pipelines for collaborative AI model development, engaging data scientists, machine learning engineers, and platform engineers.
The significance of ZenML lies in its ability to empower organizations to construct their proprietary AI models. While they may not aspire to rival a GPT-4, they can build more tailored models that cater to their specific requirements, thereby diminishing reliance on API providers like OpenAI and Anthropic.
Louis Coppey, a partner at VC firm Point Nine, notes that “the idea is that, once the first wave of hype with everyone using OpenAI or closed-source APIs is over, [ZenML] will enable people to build their own stack.”
Earlier this year, ZenML secured additional funding in an extension of its seed round, with Point Nine and existing investor Crane participating. To date, this Munich-based startup has amassed $6.4 million in funding.
ZenML’s founders, Adam Probst and Hamza Tahir, previously collaborated on a company that specialized in building ML pipelines for a specific industry. They realized the need for a modular system that could adapt to diverse circumstances, environments, and customers, thereby eliminating redundant work and culminating in the creation of ZenML.
Simultaneously, ZenML aimed to provide a head start for engineers entering the field of machine learning through its modular system. The term “MLOps” emerged, akin to DevOps but tailored for machine learning.
ZenML’s core concept revolves around pipelines. These pipelines can be written, executed locally, deployed with open-source tools like Airflow or Kubeflow, or leveraged through managed cloud services such as EC2, Vertex Pipelines, and Sagemaker. ZenML also integrates with open-source ML tools from Hugging Face, MLflow, TensorFlow, PyTorch, and others.
ZenML CTO Hamza Tahir describes ZenML as a unifying element that provides a multi-vendor, multi-cloud experience, enhancing connectors, observability, and auditability in ML workflows.
The company initially released its framework on GitHub as an open-source tool and has garnered over 3,000 stars on the platform. ZenML has also introduced a cloud version with managed servers, with CI/CD triggers in the pipeline.
ZenML has found application in various industrial use cases, including e-commerce recommendation systems, image recognition in medical settings, and more, with clients such as Rivian, Playtika, and Leroy Merlin.
The success of ZenML hinges on the evolution of the AI ecosystem. Currently, many companies rely on OpenAI’s APIs for AI features, but they face challenges such as complexity and high costs. ZenML believes that the market will trend toward creating specialized, cost-effective models in-house, driven by open-source solutions.
Sam Altman, CEO of OpenAI, also acknowledges the need for a hybrid approach, combining specialized models with broad models. Additionally, evolving regulations, particularly in Europe, could encourage companies to use AI models trained on specific datasets and in specific ways.
As Hamza Tahir puts it, “The value of MLOps is that we believe that 99% of AI use cases will be driven by more specialized, cheaper, smaller models that will be trained in-house.”
