Kubernetes AI Day North America 2022
2022-10-25
Detroit, United States
[Kubernetes AI Day North America 2021, Kubernetes AI Day North America 2022]
who it's for
machine learning engineers, data engineers, platform engineers, MLOps engineers, site reliability engineers, data scientists
topics
Kubernetes model serving with KServe; feature store architectures (Feathr/Feast); Argo Workflows orchestration; Ray on Kubernetes with JupyterLab; GPU utilization monitoring and reclamation; model drift detection and explainability; secure model artifact handling and dependency scanning; batch processing platforms on Kubernetes; data locality and CSI Alluxio mounts; model distillation for low-latency inference; CI/CD for ML with Kubeflow and Argo CD; model registry and reproducible experiments (MLflow)
| title | author | length |
|---|---|---|
| Welcome + Opening Remarks: The State of Production MLOps in the Cloud Native... | Alejandro Saucedo | 39m |
Argues that the ML lifecycle begins at deployment and lists concrete controls: an architectural blueprint separating training, artifact store and inference… | ||
| Keynote: Building Enterprise AI/ML Platform on Kubernetes | Maulin Patel | 12m |
Organizations build ML platforms on Kubernetes for four reasons: portability, customizability, performance, consistency, claiming Kubernetes supports… | ||
| Flawed ML Security: Mitigating Security Vulnerabilities in Data & Machine... | Adrian Gonzalez-Martin | 28m |
Model serving surfaces expose concrete failures: pickle deserialization lets a loaded model run arbitrary system calls, pip installers can execute code at… | ||
| How Bloomberg Uses AI to Help Speed up Over-the-Counter Trading | Camilo Ortiz, Philipp Meerkamp | 24m |
Bloomberg deploys a pipeline that detects negotiated security offers inside group chats and emails, using Kubernetes CRDs plus KServe and Kubeflow for… | ||
| Are You Really Out of GPUs? How to Better Understand Your GPU... | Natasha Romm, Raz Rotenberg | 32m |
Shows how to spot real GPU waste with concrete metrics and tools: scrape per‑GPU utilization from dcgm_exporter, query Prometheus to find GPUs with utilization… | ||
| Building a Batch Processing Platform... | Rakesh Subramanian Suresh, Aroop Maliakkal Padmanabhan | 29m |
Intuit builds a multi-cluster batch platform using Argo Workflows, Argo Events and Argo CD plus Kafka to run ~40–50k pipelines with thousands of concurrent… | ||
| Building a Data Science Platform with Argo Workflows | J.P. Zivalich, Will Wang | 27m |
Describes a working data‑science control plane that uses Argo Workflows plus a small agent per customer cluster to schedule across many clusters, a hosted… | ||
| Next-Generation Data Science Workflows Using Ray | Erik Erlandson | 31m |
Shows Ray running on Kubernetes with JupyterLab via Open Data Hub and OpenShift: a tiny self-service workflow creates a Ray cluster CR from a notebook using a… | ||
| On Data Locality in Kubernetes | Chen Wang, Shouwei Chen | 29m |
Built a Tesseract-based benchmark on three 8‑CPU 32GB worker nodes to measure data locality effects, showing up to 10x local vs remote throughput for reads and… | ||
| DevOps in Data Science: What Works and What Doesn’t | Chase Christensen, Stefano Fioravanzo | 22m |
An engineering team builds a Kubeflow-based CI/CD pipeline using Kaniko + Kubeflow Pipelines to let data scientists self-build containers and uses ArgoCD for… | ||
| Closing Remarks | Yuzhui Liu | 04m |
Conference wraps after 60 submissions narrowed to the day’s lineup, with thanks to sponsors including Google Cloud and erector, and a note that Kubernetes ML… | ||
| Feathr, an Open Source, Battletested, Scalable and Enterprise-Grade Feature Store | Hangfei Lin | 33m |
Open-source feature store built from five years of LinkedIn production use, integrated with Azure and accepted into the Linux Foundation AI & Data sandbox… | ||
| nothing matches | ||