PodcastOn Structuring an ML Platform 1 Pizza TeamPushed backThe platform team's size was described as four technical people earlier and as five people later in the discussion.31:59
Threads / The platform team / 2022
The platform team in 2022
31 sessions
PodcastOn Structuring an ML Platform 1 Pizza Team
PodcastData Mesh: Data Quality Control Mechanism for MLOps?ClaimA self-service data platform should let data producers focus on producing data products instead of managing infrastructure.12:30
PodcastPlatform Thinking: A Lemonade Case StudyPushed backLemonade's platform does not require one fixed level of openness: experienced users can customize decisions while other users can rely on defaults.27:21
PodcastPractitioners Guide to MLOpsClaimDonna Schut and Christos Aniftos create customer solutions by gathering customer requirements, piloting solutions, improving them with internal input, and publishing them after repeated success.7:18
MeetupTrustworthy Data for Machine LearningClaimConvoy’s data platform team owns data infrastructure from instrumentation and ETL through Snowflake, orchestration, data discovery, experimentation, machine learning deployment, and a feature store.9:34
Reading groupFeature Stores at Shopify and SkyscannerPushed backMatt Delacour says Shopify's internal library is a thin layer on top of Feast, rather than a copied subset of Feast.20:24
PodcastLessons from Studying FAANG ML SystemsPushed backErnest Chan argued that teams should not simply copy the priorities of large companies because each platform reflects its own company’s requirements.32:51
PodcastBuilding ML/Data Platform on Top of KubernetesClaimBuilding a machine learning platform from scratch takes about two years, even when the builders already know what they are doing.7:51
PodcastML Platform Tradeoffs and Wondering Why to Use ThemPushed backThe idea that a platform should force every team to use it was rejected in favor of making the platform appealing.20:59
PodcastA Journey in Scaling MLPushed backGabriel Straub takes an intermediate position on whether teams should use machine learning, recommending a simple rule-based baseline before adopting machine learning where it adds value.37:50
PodcastBringing Audio ML Models into ProductionPushed backValerio Velardo disputes the idea that a centralized machine learning platform should be designed completely at the beginning; he says it must evolve as specific needs emerge.46:41
PodcastThe Shipyard: Lessons Learned While Building an ML PlatformPushed backThe platform team preferred the generic serving request and response format, while the ML engineers wanted a purpose-built format for news-article extraction.18:11
PodcastMLOps as Tool to Shape Team and CultureClaimCiro Greco says that the main challenge in adopting a tool stack across a larger organization is organizational change, including security concerns, unfamiliar tools, and team skepticism.26:30
PodcastFastAPI for Machine LearningClaimBuilding FastAPI required understanding complex Python internals and studying several API-related standards.11:31
PodcastDeclarative Machine Learning Systems: Big Tech Level ML Without a Big Tech TeamPushed backPiero argued that most organizations should use an existing machine learning platform rather than build the entire training and deployment stack themselves.48:10
PodcastMaking MLflowPushed backThe speakers differed over whether MLflow should remain focused on independent components or move toward a more complete end-to-end platform.26:30
PodcastWhy and When to Use Kubeflow for MLOpsClaimKubeflow provides exploration, pipeline orchestration, and serving tools on an open-source platform.6:50
PodcastMLflow Pipelines: Opinionated ML Pipelines in MLflowClaimXiangrui Meng stayed at Databricks because its founders had a clear vision of cloud computing and a unified analytics platform, and executed consistently toward that vision.5:30
PodcastBuilding Better Data TeamsPushed backLeanne Fitzpatrick preferred building the development and deployment stack more in-house and openly, rather than adopting a vendor stack that would force a major team-wide change.43:49
PodcastHow Hera is an Enabler of MLOps IntegrationsClaimDyno Therapeutics treats machine learning as an internal innovation engine for improving its biological research rather than as an external-facing product.17:49
PodcastScaling Machine Learning with Data MeshClaimA self-service machine learning platform can begin with reusable blueprints, a command-line interface, infrastructure provisioning, and repository templates before adding a graphical interface.28:49
PodcastML Platforms, Where to Start?ClaimOlalekan Elesin joined Scout24 as a data landscape engineer and later moved into technical product management for the AI platform.3:19
PodcastMLOps at DoorDashClaimThe DoorDash machine learning platform aims to cover the applied machine learning lifecycle end to end and at scale, from feature engineering through model serving.4:23
PodcastBringing DevOps Agility to MLClaimLuis Ceze describes OctoML as a machine learning deployment platform intended to bring DevOps agility to machine learning deployments through hardware independence, automation and performance.27:26
PodcastDatabricks Model Serving V2Pushed backKubernetes is not always the right platform for machine learning; its suitability depends on organizational maturity, scale, and available support.15:10
MeetupDriving ML Data Quality with Data ContractsClaimGoCardless has an internal fraud model and two customer-facing models called Success+ and Protect+.4:52
Podcast"Real-Time" ML: Features and InferenceClaimSasha Ovsankin says the value and return on investment of real-time features were initially unclear at LinkedIn, so the team adopted a gradual approach and evaluated specific use cases.25:41