As more and more people use ChatGPT and other AI-driven tools, companies face more pressure to integrate language models and machine learning into their products. Increasingly, executives recognize that they must leverage ML in their products and services to remain competitive. Moreover, software and ML engineering teams are expected to deliver increasingly complex ML-driven applications that use bigger models and real-time data.  

However, although AI research and models have made significant progress in recent years and ML teams face higher expectations; AI tools and platforms are just beginning to advance. As a result, data and ML teams need more developer-friendly tools to enable them to iterate quickly yet responsibly. They need tools like Metaflow.  

Metaflow is an open-source human-centric Python framework created by the Outerbounds team during their tenure at Netflix. Although Netflix had developed a platform for data scientists and ML engineers to build recommendation models, this system was not flexible enough to support the agile development of other projects ranging from causal machine learning to model-based anomaly detection. As such, Ville Tuulos, Savin Goyal, and their team developed Metaflow to help data scientists and ML engineers easily build and manage projects (ranging from operations research to NLP) without expending effort on versioning, dependency management, or compute resource management. The framework enables users to easily move from running an ML/DS pipeline on a local machine to deploying on cloud resources.

Specifically, Metaflow helps teams speed up prototyping and deploy promising prototypes to production quickly, release more data science applications with fewer resources, and make applications more robust by enforcing best practices for building and operating production applications. It achieves these goals by providing a unified API to the infrastructure stack as well as tools for model version control and experiment tracking. 

Although several new MLOps vendors have emerged in the past few years, Metaflow stands out for three reasons:

• Product: Unlike many other ML tools and platforms, Metaflow is human-centric. It provides user-friendly interfaces to build and deploy data science workflows so data scientists can focus on R&D and not infrastructure. It’s a versatile tool for data and ML practitioners and helps other stakeholders collaborate on data science projects. In addition, it can handle large-scale ML workflows, so it’s suitable for large enterprises with significant amounts of data. It also includes version control and reproducibility features to help enterprises meet compliance and governance requirements. 
• Team: The Outerbounds founding team members have dedicated their entire professional careers to building better tools for data/ML practitioners. Ville previously co-founded a startup focused on building a novel scriptable data platform and subsequently developed an in-memory data store for analytics as a Principal Engineer at AdRoll. Savin built the ML infrastructure that powered real-time bidding at RocketFuel and the Economic Graph at LinkedIn. Oleg was a founding engineer at Alpine AI, building a natural language understanding platform, and worked as a SWE at Tecton, which makes one of the most popular feature stores. This team will not rest until data and ML practitioners have better tools. 
• Community: The Outerbounds community continues to expand as more and more teams standardize on Metaflow and engage with each other through the company’s Slack channel. Their Slack channel has become a popular resource for ML practitioners to swap notes on best practices for designing and operating ML stacks. Participants regularly share advice on how to integrate with data sources and compute environments, accelerate time to market by simplifying the deployment and management of data science projects, and more. 

For these reasons, we’re also delighted to announce our investment in Outerbounds and excited to support them as they announce the release of the Outerbounds platform, which will enable teams to deliver ROI even faster by shortcuting months or even years of infra. We are so grateful for the opportunity to continue to partner with them as they empower ML teams to deliver better ML projects faster.