Cake “cut a year off the product development lifecycle” for a leading R&D platform in the materials science industry.
Cake delivers the equivalent value of three senior engineers by deploying and managing full stack AI/ML infrastructure.
Robust tenant isolation and comprehensive security ensure traditional companies can adopt new AI/ML technologies.
Platform configurability and accelerated time-to-market made Cake the “extremely obvious” choice as a new alternative to the traditional "build" and "buy" options.
For chemical scientists, many problems are physically and computationally impossible to brute force. The traditional scientific method – develop a hypothesis, design tests in the labs, derive the next experiment, and repeat – is highly inefficient for multivariable optimization.
Machine learning offers the potential to empower materials science researchers and revolutionize the chemical industry through rapid acceleration of material discovery and innovation. Virtual experiments built on machine learning models can deliver millions of iterations several orders of magnitude faster and cheaper than working in a laboratory, surfacing the best candidates for further physical experimentation and prototyping.
A Cake customer (which has asked to remain anonymous) is a Series A tech startup dedicated to building a new R&D platform for the chemical industry. This company empowers chemists with tools and workflows that make powerful machine learning capabilities accessible to everyday chemists. Their product helps the largest chemical manufacturing companies in the world reduce time to market, streamline experiment complexity, and eliminate the material costs of unnecessary experimentation.
This leading R&D platform selected Cake to manage its machine learning and AI infrastructure.Working together, Cake provides the technical foundation to train hundreds of models securely and provides tools that scale to thousands of users executing data science workflows concurrently.
Now, each time a chemist wants to investigate an inverse design problem, they can build, tune, and deploy their own models and run millions of virtual experiments using those models as endpoints. Shifting to virtual experimentation has been proven to save materials science companies tens of thousands of dollars and months of development per experiment.
The majority of companies supported by this R&D platform are many decades old and demand stringent data security and privacy requirements from the platform. In some cases, this R&D platform is the first cloud system these companies have ever brought into their research organizations.
As this customer's AI infrastructure platform, Cake provides comprehensive security and access controls, including:
Fine-grained RBAC for different services and secure authentication via SSO
Configuration of IAM roles and permissions for CI/CD
End-to-end encryption with advanced networking policies and security controls
Secure access to private container registries
With multiple chemical manufacturers on this customer’s platform, Cake has also implemented robust tenant isolation with multiple isolated environments for development and production. Cake is able to fully support this customer’s R&D platform without having access to its production environment, ensuring customer data remains strictly protected.
Example architecture for tenant isolation within Cake
“Cake manages all of our base infrastructure and ensures everything is up to date. With Cake, there’s so much we don't have to worry about while getting the performance and features we need.”
Head of Data Science, Materials Science R&D Platform
Given the breadth of these virtual experiments, this customer’s R&D platform needed flexible computing resources to dedicate to long-running jobs, with the ability to scale as needed. The size and technical complexity of the operation could not be deployed in a traditional containerized service on its own.
Working with Cake, the team implemented autoscaling for GPU and CPU workloads for their platform infrastructure. Cake was then used to deploy, integrate, and maintain open source AI components across their stack. For predictive ML, this stack included:
ClearML - experiment tracking
Kubeflow - pipelines and notebooks
Ray Train - model training
KServe - inference server
Ray Serve - inference server
This company has also explored generative AI workflows for chemistry modeling using a stack that included:
Llama 3 models (both 8B and 70B parameter variants)
vLLM - inference server
DSPy - prompt engineering in AI pipelines
Milvus - vector storage within ML operations
OpenTelemetry - observability
Components implemented with Cake also enable the team to closely monitor models for drift and performance.
Cake is a core platform we can customize however we want. We needed that level of control. We can't just go with a standard off-the-shelf deployment platform – they are not customizable and don't work for the complexities of the problems we are solving
Head of Data Science, Materials Science R&D Platform
The R&D platform team faced a set of compromises when considering their approach. SaaS products offered turnkey solutions but would not be flexible enough and might jeopardize information security. Outsourcing the build to a third party would save time but would create a problem over time in maintenance and support. The team did not have sufficient resources internally, and hiring additional engineers would take months, with a significant risk of scope creep for additional required skills.
As a fully managed and highly configurable platform, Cake offers an elegant solution. As this customer’s Head of Data Science described, “We can build anything we want on top of the Cake base infrastructure - and then we also benefit from the feedback and different features driven by other customers.”
Cake provided the R&D platform team with a secure method of building a long-term sustainable product. As their CEO explained, “The biggest thing came down to ownership. Cake stands up in our infrastructure, allows us to maintain all the information security standards across our organization, and is supported as we have new needs.”
He continued, “For a company our size, it's frankly a no-brainer. The alternative is to spend more money to get a worse product. It's extremely obvious to build with Cake.”
“Cake cut a year off our product development cycle. That’s the difference between life and death for small companies.”
CEO, Materials Science R&D Platform company
By bringing in Cake for infrastructure management rather than hiring a larger platform engineering team, this customer’s R&D group was able to drive significant resource savings. According to their CEO, “Without Cake, we would need three highly-capable senior engineers to maintain our infrastructure - not to mention the sunk cost of setting the whole thing up.”
He continued, “Cake is half the cost of hiring a single FTE, and you get results faster."
As their team described,the greatest value of working with Cake was found not in the salary savings but rather in the accelerated time to market. Cake’s turnkey infrastructure saved the platform team a year of development. As their CEO detailed, “It's not just the cost saving – we just turned it on, and it worked. The time savings is worth way more than three engineers' salary at the end of the day.”
Time savings will also be an ongoing benefit through maintenance and upgrades to the Cake platform. As new components and technologies emerge, they are incorporated into Cake and are made available to customers. Their Head of Data Science explained, “With Cake, everything works together beautifully. It's so streamlined. There are many moving pieces whenever there’s an update, but Cake just works.”
Working with Cake, this team saved an entire year of product development time. To learn more about how Cake helps teams launch machine learning and AI applications quickly, please sign up for a demohere.