11-11, 14:00–14:30 (MST), Theater
Your team wastes hours each day constructing sequences of operations to accomplish tasks. Foyle is an AI that translates an operator’s high level intent (e.g. deploy this python code on a GPU) into a sequence of operations (e.g. kubectl commands). Foyle learns this translation automatically, reducing the need for platform teams to create handcrafted tools.
Foyle is an open source AI (foyle.io). To train itself, foyle uses vscode and RunMe.Dev to make playbooks executable. This UX allows Foyle to log intents, actions and human feedback. Using this data, Foyle continuously improves its ability to translate intent into operations.
In this talk you will learn: 1) how we model operations as a sequence prediction problem perfect for LLMs 2) how we create a UX that logs implicit human feedback and 3) how we use this feedback to retrain the model.
This talk will show platform engineers how to use AI to build the next generation of platforms and not just enable AI for others.
Building a transformational DevOps copilot will require cooperation across organization and projects. Foyle.io is one piece of a much larger stack. Foyle relies on open source projects like vscode, RunMe.Dev and continue.dev to create the right UX for interacting with an assistant. Other projects like open-flux-ai and k8sgpt are critical for building AIs that are experts in different parts of the CNCF landscape. This talk is an opportunity to connect with those other teams and build a strong community to pursue the goal of using AI to improve velocity and minimize outages.
This talk will give the audience a concrete understanding of how AI can address the problems platform teams are trying to solve. This talk doesn’t treat AI as magic dust to be sprinkled everywhere. By understanding the difficulty of dealing with Cloud we can frame the problem in a way that is well suited to the capabilities of LLMs. Platform teams will learn that their domain expertise makes them uniquely suited to begin to use AI to solve these problems.
I’m a Machine Learning platform engineer with over 15 years of experience. I’m an expert in using AI to solve practical business applications. I’ve built platforms for YouTube, Google Cloud Platform, and Primer to enable ML Engineers and data scientists to rapidly develop and deploy models into production. I played a pivotal role in developing systems like YouTube’s Video Recommendations. I also made major contributions to open-source software, including creating Kubeflow, one of the most popular OSS frameworks for ML.
I’m an expert in Cloud services, Kubernetes, MLOps, LLMOps, CICD, IAC, and GitOps.
I’m currently developing an AI assistant, Foyle (www.foyle.io) to simplify deploying and operating software.
I live in the Bay Area with my dog, Teddy. When I’m not working, I enjoy skiing, baking chocolate chip cookies, and hanging out with Teddy.