ZenML is an open-source framework that aims to connect and unify various open-source AI tools. It allows data scientists, machine-learning engineers, and platform engineers to collaborate and build new AI models through pipelines. One of the key advantages of ZenML is its ability to empower companies to build their own private models, reducing their reliance on API providers like OpenAI and Anthropic.
According to Louis Coppey, a partner at VC firm Point Nine, ZenML will enable people to build their own AI stack once the initial hype of using OpenAI or closed-source APIs subsides.
ZenML recently raised funding from Point Nine and existing investor Crane, bringing its total funding to $6.4 million. The Munich-based startup was founded by Adam Probst and Hamza Tahir, who previously worked together on ML pipelines for a specific industry. This experience led them to develop a modular system that could adapt to different circumstances and customers, ultimately resulting in ZenML.
ZenML is particularly useful for engineers starting with machine learning, as it provides a modular system that can be easily customized. The ZenML team refers to this space as MLOps, similar to DevOps but specifically focused on ML.
The core concept of ZenML is pipelines, which can be run locally or deployed using tools like Airflow or Kubeflow. It also integrates with managed cloud services such as EC2, Vertex Pipelines, and Sagemaker. Additionally, ZenML integrates with popular open-source ML tools like Hugging Face, MLflow, TensorFlow, and PyTorch.
ZenML aims to bring together various tools and provide a unified experience for users. It offers connectors, observability, and auditability to ML workflows. Initially released as an open-source tool on GitHub, ZenML has gained significant popularity with over 3,000 stars on the platform. The company has also launched a cloud version with managed servers, and CI/CD triggers will be available soon.
ZenML has been used by companies in industrial use cases, e-commerce recommendation systems, and medical image recognition. Clients include Rivian, Playtika, and Leroy Merlin.
The success of ZenML will depend on how the AI ecosystem evolves. Currently, many companies rely on APIs like OpenAI, but there are issues with their sophistication and cost. ZenML offers a more cost-effective and customizable solution, which is appealing to companies. Additionally, the increasing focus on ethics and regulations in AI usage could encourage companies to use specialized models trained on specific data sets.
ZenML aligns with the belief that the majority of the market will require its own AI solutions, rather than relying solely on large language models built by companies like OpenAI. The future is expected to be a hybrid of both broad models and specialized, smaller models.
Overall, ZenML aims to usher in the next phase of AI production, where specialized models trained on proprietary data become more prevalent. With 75% of enterprises shifting from proofs of concept to production by 2024, the value of MLOps becomes even more significant in driving AI use cases.
