Today, we are announcing a major drop of new features in the Outerbounds Platform. To get a quick overview of the platform including the new features, watch this product walkthrough:
Since 2019, we have been helping hundreds of companies from startups to Fortune 500s ship data, ML, and AI projects faster, developed by happier scientists and engineers using open-source Metaflow. Their use cases have been exhilaratingly diverse, from forecasting and fraud detection to computer vision for self-driving, drones, and cancer research.
Over the past months, the interest towards LLMs and Generative AI has exploded, greatly growing the desire to integrate data, ML, and AI in various systems. While the adoption of LLMs and generative AI for production use cases is still nascent - and will remain so for many months to come - it has prompted companies to question whether their existing ML and data initiatives are aligned with the upcoming needs around the new AI use cases.
Many of our recent customers have chosen the Outerbounds Platform to future-proof their existing data engineering pipelines, traditional ML models, and to up level their teams today, anticipating rapid growth in their data, ML, and AI initiatives.
Convoy, a leading digital freight network in the US, is a great example of this, as highlighted by Marc Millstone, a Principal Engineer Manager at Convoy:
The Outerbounds Platform empowers ML scientists to quickly scale our model training and less technical users to easily solve “mundane” analytical tasks in a clean and repeatable manner. By unifying our data and ML pipelines we can now automatically train models when an underlying warehouse table is updated as opposed to relying on manually scheduled independent pipelines, which drastically improves efficiency and collaboration.
Four new features to support serious data, ML, and AI teams
Today, we are releasing four new features in the Outerbounds Platform to support companies that treat data, ML, and AI as a core component of their business. The features have been developed in collaboration with a number of our customers to address their immediate needs, as well as anticipate the needs of tomorrow.
For years, we have been developing Metaflow with the goal of making it the most user-friendly and productive way to develop data, AI, and ML apps, and deploy them to production quickly. Today's features align with this story:
Let's take tour of the new features:
- Built-in cloud workstations.
- Develop custom Generative AI and large language models with flexible GPU resources.
- Support for building reactive ML and AI systems.
- Bank-grade security and compliance.
Built-in cloud workstations
One of the superpowers of Metaflow has always been first-class support for local development and testing. However, this assumes that you
- have your laptop or other development environment configured properly to access the company’s cloud environment securely,
- have all necessary dependencies installed, notebooks alongside a modern IDE, and
- have enough horsepower - often including GPUs - to develop effectively.
These are not simple requirements! Many organizations fail to supply their data scientists and ML engineers with these foundational tools of the trade. With Outerbounds cloud workstations, data developers get the best of local IDEs and modern cloud-based development environments, optimized for security-conscious, data-intensive work:
- In contrast to other cloud workstations like GitHub Codespaces, these workstations run on your cloud account, so no data or processing leaves your account, which is a requirement for many privacy-sensitive companies.
- Since the workstations run on your account, they can be configured to leverage any instances that match your use cases and reside close to your data, resulting to significantly better performance for data-intensive development at lower cost.
Also, the Outerbounds workstations
- work seamlessly with local VSCode IDEs (no need to develop only in the browser),
- support Jupyter notebooks and all Metaflow functionality out of the box,
- provide a unified, secure, centrally managed, consistent environment - no more fighting with library dependencies and inconsistent configuration, and,
- handle authentication and authorization centrally via your SSO.
Read more on our new product page for workstations.
Develop custom GenAI and LLMs with flexible GPU resources
We have been actively working with the research community to make sure that Metaflow and Outerbounds Platform work well for state-of-the-art generative AI and LLM use cases. We believe that over the coming years, many companies will want to retain control over their data and models, differentiating their offerings from unrefined foundation models and generic AI APIs.
To support them in this rapidly evolving landscape, we have been testing popular models with Metaflow and publishing popular recipes for many open-source foundation models, such as Stable Diffusion, Whisper, Dolly, and LLaMA.
In addition to the modeling layer, we have worked actively on the compute layer, making sure GPUs are easily and cost-effectively accessible to all users of the Outerbounds Platform. Besides supporting GPUs offered by AWS, GCP, and Azure, we have partnered with CoreWeave, one of the largest dedicated GPU clouds, to expand the pool of available GPU resources that are available to our customers easily. Also, we are happy to help you to leverage local GPU clusters.
Imagine executing your code on an instance with four A100 GPUs simply by writing a line of code, @resources(gpu=4), paying a low price only for the seconds the code executes. Behind the scenes, the Outerbounds Platform takes care of the heavy lifting of allocating the hardware resources, shipping the code and its dependencies for execution securely, and tracking progress and results.
Expect to see many more announcements on this front during the course of the year! Meanwhile, you can read more at our new product page for LLM, GenAI, and GPUs.
Building reactive, event-based systems
While the new workstations allow you to develop and experiment effectively, the new built-in support for reactive workflows allows you to build sophisticated production systems powered by ML, integrating workflows to data warehouses like Snowflake, Databricks, or open-source data lakes (e.g. based on Apache Iceberg), as well as your other systems downstream, similar to the system that powers all business-critical data processing and data science at Netflix.
The Outerbounds Platform covers the end-to-end process of developing ML and data systems, from the simplest models to the state-of-the-art systems composed of tens of separate flows, connected together via events. You can
- Develop and test projects rapidly with secure Outerbounds Workstations, even testing events locally.
- Deploy projects as A/B experiments or to production through CI/CD systems, running in a highly-available, scalable environment.
- Connect the projects to surrounding systems through real-time events, allowing them to benefit from ML and AI seamlessly.
The event-triggering feature that enables this has been available in open-source Metaflow since the recent 2.9 release, and now the Outerbounds Platform makes the feature enterprise-ready:
- The platform comes with a built-in, highly available event bus, allowing it to handle a massive volume of events reliably.
- External event sources can be authenticated with centrally-managed machine-to-machine tokens, making it easier to integrate real-time ML to external systems securely.
- The platform comes with a growing library of integrations to Snowflake, Databricks, and other data systems.
Read more in our new product pages for data access, production workflows, and first-class experimentation.
Bank-grade security and compliance
Thanks to our roots in large business-critical use cases, we have wanted to make sure that ML and data science teams working in highly regulated industries can benefit from the Outerbounds Platform. We have been working with a number of fintechs to ensure that their requirements are being met. A great example of this is a major European fintech, Trade Republic:
We are a bank, everything we do needs to be auditable. This means we need to be able to reproduce what was in production and the Outerbounds Platform gives us that for free, as all models and metadata are versioned. I sleep much more comfortably knowing this.
- Thanasis Noulas, VP of Data Science, Trade Republic
To support the compliance requirements of customers in highly regulated industries, the Outerbounds Platform comes with a number of critical security features:
- Deploys in your cloud account with a tight security perimeter.
- Authentication and authorization integrates with your SSO system, such as Okta, Google, or Azure Active Directory.
- Machine-to-machine tokens for programmatic authorization.
- Integration with centrally managed secrets.
- Complete audit trail of all user and infrastructure events.
- SOC2 Type 2 compliance.
In addition, the platform can be deployed to most geographic regions to comply with data residency requirements. Read more in our new product page about security features.
Not forgetting our human-centric focus
Ever since the initial open-sourcing of Metaflow in 2019, its key differentiator has been a delightful user experience. All APIs are carefully crafted to fit in the hands of data-centric developers and engineers supporting them, making sure developers with diverse domain expertise can focus on actual business problems, not on infrastructure.
In addition to the product design, we invest heavily in support and education. This is particularly important now that there is a lot of excitement, but also confusion and doubt, about the future of ML and AI. Over the past months, we have greatly expanded our available learning paths:
- If you prefer learning hands-on, our free, interactive sandbox environment for Metaflow was recently upgraded to contain a ton of interactive hands-on tutorials and content, reflecting much of the functionality available in the Outerbounds Platform.
- If you prefer reading, we provide information about many topics in the Resources for Data Scientists, open-source Metaflow’s comprehensive documentation, or you can buy a book about Metaflow.
- If you prefer watching, take a look at our popular Fireside Chats series on YouTube.
- If you prefer learning with a live instructor, you can sign up to a highly-rated course about Metaflow and become a certified Metaflow expert.
- If you prefer learning by asking questions, join thousands of other data scientists and engineers on the active Metaflow community Slack.
In addition to these resources, organizations on the Outerbounds Platform get access to onboarding customized for the needs of their teams, regular office hours, and dedicated enterprise support. The platform is fully managed 24/7 with a guaranteed SLA so both engineering teams as well as data developers get a solid foundation for their work.
Start your free trial in 15 minutes
We know that data and ML teams, as well as engineering teams supporting them, are short-staffed and busy with existing responsibilities. Having this in mind, we have made it extremely easy to get started with the platform. It takes about 15 minutes of engineering time.
If you are curious to see how the Outerbounds Platform could work for you, schedule a quick call and we can get you started with a free trial today!
