MeetupTeam Aurora: Accelerating ML with KubeflowClaimAurora separates compute from machine-learning orchestration so workloads can run on different systems, including internal systems and SageMaker.9:18
Threads / The platform team / 2023
The platform team in 2023
19 sessions
MeetupTeam Aurora: Accelerating ML with Kubeflow
PodcastMachine Learning CareClaimMachine-learning products require more than engineering and science; they also involve go-to-market work, customer-facing work, and internal project management.0:47
PodcastML in Production: A DS from Ubisoft PerspectivePushed backJean-Michel Daignan rejects the idea that data scientists should be required to adopt extensive unit and integration testing, emphasizing scalability testing instead.14:54
PodcastAirflow Sucks for MLOpsPushed backStephen Bailey argues that data teams should balance business value and platform quality rather than always building the most rigorous solution before releasing anything.19:39
PodcastML Scalability ChallengesClaimWaleed Kadous predicts that high-value or batch-oriented use cases will be adopted before lower-value real-time use cases because of the cost of large language models.39:04
PodcastMachine Learning Education at UberPushed backMelissa Barr said that a company without a homegrown machine learning platform may not need to build its own internal education program because strong external resources already exist.42:26
PodcastMLOps Build or Buy, Startup vs. Enterprise?ClaimSlack replaced an external model-serving service with internal infrastructure because integrating the external service with Slack's infrastructure and policies was too difficult.36:41
PodcastWhy is MLOps Hard in an Enterprise?ClaimMaria Vechtomova says Model Factory and later versions of her MLOps framework were intended to create a golden path so data scientists could bring models to production without rebuilding the same process each time.9:09
PodcastClean Code for Data ScientistsClaimMatt Sharp says Shopify's Merlin is a machine learning platform that helps data scientists manage machine learning projects.30:23
PodcastEliminating Garbage In/Garbage Out for Analytics and MLClaimRoy Hasson says that product teams should talk directly with users to understand their needs, problems and feedback instead of only building from internal assumptions.13:27
LLM on KubernetesPushed backRahul Parundekar argues that companies should prioritize a repeatable deployment platform and model iteration over over-optimizing Kubernetes autoscaling while GPUs are scarce.16:42
PodcastFrom Virtualization to AI IntegrationPushed backThe early cloud virtualization history and the adoption timeline for VMware are uncertain, with Lamia Youseff explicitly saying she might be wrong about the details.16:53
PodcastBuilding an ML Platform: Insights, Community, and AdvocacyPushed backStephen Batifol disputes the idea that Kubeflow is a good fit for his platform, saying that the main problem is its execution and developer experience rather than the overall concept.9:08
PodcastMLOps at GetYourGuidePushed backJean Machado argued that people should not be blocked from using LLMs because they do not use the platform's Python-centered templates, although the platform will provide managed tooling for use cases that need observability and controls.54:11
PodcastThe Future of Feature Stores and PlatformsClaimFeature stores and feature platforms are different concepts, and a feature platform includes a feature store plus broader workflows for production machine learning.4:21
MeetupScaling MLOps for Computer VisionClaimDavid Espejo says Flyte is an open-source, machine-learning-aware continuous delivery platform intended to bridge model developers and operations teams.10:41
PodcastChallenges Operationalizing ML (And Some Solutions)Pushed backNathan Ryan Frank rejects the assumption that every available ML tool should be adopted, arguing that teams should first identify the problem and assess whether the tool provides a compelling, easy outcome.15:33