Stevanato Group, a global provider of integrated containment and delivery solutions to the biopharmaceutical and life sciences industries, has launched an Artificial Intelligence (AI) platform, based on Deep Learning (DL) models, that leverages the benefits of human-like decision-making in automatic visual inspection equipment. The platform allows pharma companies to overcome the traditional trade-off between detection rate and false rejection rate, ensuring robust and replicable results. False rejects can be reduced tenfold, and the detection rate can be improved, yielding up to 99.9% accuracy, both for particle inspection and cosmetic defects detection.

Deep Learning is particularly beneficial when applied to difficult-to-inspect and high-value biotech drugs, as it allows pharma companies to achieve higher productivity while preserving drug integrity. Drugs in the form of suspensions or lyophilized cakes frequently challenge available vision tools, causing misinterpretations of supposed defects. Traditional systems can, for instance, misclassify cosmetic defects or air bubbles as particles. Artificial Intelligence mitigates misclassification and reduces costly re-inspection.

Thanks to its partnership with Microsoft and the adoption of Microsoft Azure platform, Machine Learning and AI features, Stevanato Group intends to deliver “smart” equipment compliant with strict pharmaceutical data management and security requirements, while improving inspection performance and reducing costs related to production reparametrization. The certified cloud-based platform is compliant with U.S. CFR 21 Part 11 and EU GMP Annex 11, meets data integrity needs, and offers advanced monitoring tools such as heat maps and confusion matrix for model performance evaluation.

Incorporating new technologies can be challenging for pharma companies, as it usually requires them to adapt internal processes and invest time and resources. The combination of the latest digital developments and a qualified team of vision and AI engineers bring a dedicated and accurate data analysis service with continuous support for tasks from image collection to model validation.

“We are glad to provide our customers with inspection machines equipped with the latest AI technology, which are able to substantially increase defect detection accuracy even with the most challenging drugs,” said Raffaele Pace, engineering vice president of operations, Stevanato Group. “Choosing Microsoft Azure as our technology platform allows us to offer a proven solution while delivering the highest standards in terms of data security. Thanks to our skilled visual inspection engineers and experts in big data analysis, we can support clients from the implementation of Artificial Intelligence to ongoing support during operations.”

Key features of the new AI platform include:

• Cloud-based: Data remains online and therefore perpetually available. The certified cloud-based platform that stores images and data can work with any cloud-based system and allows operators to manage images even if they are stored in the server.

• Security: Compliant with U.S. CFR 21 Part 11 and EU GMP Annex 11, the platform enables data sharing in a completely safe environment. Further, multifactor authentication and encrypted communication ensure complete access control and data security.

• Assistance: Continuous support is available for all platform capabilities throughout the process, assisting pharma companies with a variety of tasks including a labeling assistant tool to optimize timing for classification and new recipe development.

• Monitoring: The platform features a wide range of statistics and visualizations (heat maps, con-fusion matrix, etc.) for model performance evaluation. Pharma companies can track and moni-tor all processes through real-time reports.

What Is Cloud AI Platform Pipelines?

Google announced the beta of Cloud AI Platform Pipelines, a managed service designed to help data scientists and ML engineers deploy robust, repeatable machine learning pipelines on Google Cloud. The service adds monitoring, auditing, version tracking, and reproducibility to ML workflows that would otherwise be difficult to operationalize at scale.

The core motivation is familiar to anyone who has moved an ML project from experimentation to production. As Ramesh and Unruh wrote in the original blog post:

When you’re just prototyping a machine learning (ML) model in a notebook, it can seem fairly straightforward. But when you need to start paying attention to the other pieces required to make an ML workflow sustainable and scalable, things become more complex.

Cloud AI Platform Pipelines is Google’s answer to that complexity — and its entry point into the broader practice of MLOps.

MLOps: DevOps Principles Applied to Machine Learning

MLOps (Machine Learning Operations) applies DevOps principles — automation, monitoring, version control, and continuous delivery — to the ML lifecycle. A typical ML workflow subject to MLOps practices covers:

• Data preparation and validation
• Feature engineering and analysis
• Model training and evaluation
• Model deployment and serving
• Ongoing monitoring and retraining

Cloud AI Platform Pipelines is designed to be the orchestration layer that ties these stages together in a repeatable, auditable manner.

Architecture: How It Works

Cloud AI Platform Pipelines runs on a Google Kubernetes Engine (GKE) cluster and is accessible through the Cloud AI Platform dashboard in the Google Cloud Console. The service has two major parts:

• Infrastructure layer: Deploys and runs structured AI workflows integrated with GCP services including BigQuery, Dataflow, AI Platform Training and Serving, and Cloud Functions.
• Pipeline tooling layer: Tools for building, debugging, and sharing pipelines and individual pipeline components.

Workflow Orchestration with Argo

Under the hood, the platform uses Argo, an open-source, container-native workflow engine for orchestrating parallel jobs on Kubernetes. Argo executes each pipeline on individual isolated pods within the GKE cluster. This architecture allows pipeline components to leverage managed Google Cloud services such as Dataflow, AI Platform Training and Prediction, and BigQuery. Pipelines can also include steps that run large-scale GPU and TPU computation directly in the cluster, taking advantage of GKE features like autoscaling and node auto-provisioning.

Automatic Metadata Tracking with ML Metadata (MLMD)

Every pipeline run benefits from automatic metadata tracking via the ML Metadata (MLMD) library. MLMD logs:

• Artifacts consumed and produced at each pipeline step
• Pipeline parameters used during each run
• Lineage linking input and output artifacts across steps

MLMD is included as part of the TFX SDK but can also be used independently in other contexts.

Choosing an SDK: KFP vs. TFX

Users can define their pipelines using one of two SDKs:

Regardless of which SDK is used, the compilation and submission process is the same: the SDK compiles the pipeline definition and submits it to the Pipelines REST API. The REST API server then stores and schedules the pipeline for execution. Google noted in the original blog post that the two SDK experiences were planned to converge over time.

Key Features for Enterprise ML Workloads

According to the announcement, customers using Cloud AI Platform Pipelines receive:

• Push-button installation directly from the Google Cloud Console
• Pipeline versioning to track changes to pipeline definitions over time
• Automatic metadata tracking of artifacts and executions via MLMD
• Cloud Logging integration and built-in visualization tools
• Seamless integration with BigQuery, Dataflow, AI Platform Training and Serving, Cloud Functions, and other GCP managed services
• Prebuilt pipeline components for common ML workflow steps, plus support for custom components

Portability: The Kubeflow Advantage

Because Cloud AI Platform Pipelines is built on Kubeflow Pipelines — which is open-source and runs on standard Kubernetes — workloads are not strictly locked into Google Cloud. A commenter on a Hacker News thread discussing the announcement noted:

Cloud AI Platform Pipelines appear to use Kubeflow Pipelines on the backend, which is open-source and runs on Kubernetes. The Kubeflow team has invested a lot of time on making it simple to deploy across a variety of public clouds, such as AWS and Azure. If Google were to kill it, you could easily run it on any other hosted Kubernetes service.

This is a meaningful consideration for teams evaluating cloud vendor lock-in risk in their ML infrastructure decisions.

Planned Features at Beta Launch

At the time of the beta announcement, Google outlined the following features on its near-term roadmap:

• Easy cluster upgrades
• Additional templates for authoring ML workflows
• Simplified UI-based setup of off-cluster storage for backend data
• Workload Identity support for transparent, credential-free access to GCP services
• Multi-user isolation — allowing each user to control access to their own pipelines and resources within a shared cluster

Context: Google’s Broader MLaaS Strategy

Cloud AI Platform Pipelines is part of Google’s broader Machine Learning as a Service (MLaaS) portfolio, which includes Cloud AutoML, Kubeflow, and AI Platform Prediction. The service positions Google more directly against competing managed ML platforms such as Amazon SageMaker and Azure Machine Learning.

For DevOps and platform engineering teams tasked with supporting data science organizations, Cloud AI Platform Pipelines represents a significant step toward treating ML workflows with the same operational rigor as software delivery pipelines — with versioning, auditability, and reproducibility built in from the start.

Further Reading

For implementation details, refer to the official Google Cloud AI Platform Pipelines documentation and the Kubeflow Pipelines project on GitHub. The original announcement was covered in depth by InfoQ.

What Is Cloud AI Platform Pipelines?

Google Cloud Platform (GCP) announced the beta launch of Cloud AI Platform Pipelines, a managed product for automating and orchestrating end-to-end machine learning (ML) workflows. The product is built on two open-source technologies: TensorFlow Extended (TFX) and Kubeflow Pipelines (KFP), and runs on a Google Kubernetes Engine (GKE) cluster.

The core problem Cloud AI Platform Pipelines addresses is the complexity of managing the full ML lifecycle — from ingesting raw data, through model training and evaluation, to serving predictions in production — in a repeatable, auditable way.

Key Features

• Repeatable ML pipelines: Define pipeline steps once and re-execute them consistently across runs.
• Artifact and lineage tracking: Track the history, versions, and provenance of data artifacts produced at each pipeline stage using ML Metadata (MLMD).
• Monitoring and auditing: Built-in observability for pipeline runs and model versions.
• GCP service integration: Native connectors to BigQuery, Dataflow, and other GCP services.
• Managed infrastructure: Google manages the underlying GKE cluster, reducing operational overhead for teams.
• Dual SDK support: Teams can use the higher-level TFX SDK or the lower-level KFP SDK; Google has stated plans to merge both into a single TFX SDK.

Architecture Overview

Understanding how Cloud AI Platform Pipelines works requires understanding its two foundational layers:

TensorFlow Extended (TFX)

TFX is a high-level abstraction layer for defining ML pipelines. The central concept is a pipeline — an ordered series of data transformation steps called pipeline components. Data passed between components are called artifacts. TFX also defines and manages an ML Metadata (MLMD) store, enabling version tracking and lineage for both pipeline runs and the artifacts they produce.

TFX was first documented publicly in a paper Google presented at KDD 2017, which described the internal ML platform Google built to cover all phases of the ML process: data analysis, transformation, training, evaluation, and production inference. The original execution engine was Apache Beam — itself derived from Google’s internal Flume system and the technology underpinning Google Cloud Dataflow. TFX continues to use Beam for defining data-parallel operations, and now also supports Kubeflow Pipelines and Apache Airflow as orchestration engines.

Kubeflow Pipelines (KFP)

KFP acts as the orchestrator. It schedules and executes each TFX pipeline component as a pod on the GKE cluster, coordinating execution order and passing artifacts between steps. Google open-sourced Kubeflow in 2018 as an ML-specific platform for Kubernetes. Spotify later adopted Kubeflow as their standard ML platform and open-sourced their Terraform templates for cluster provisioning.

Relationship to Other GCP Orchestration Tools

Before Cloud AI Platform Pipelines, Google had recommended Apache Airflow (and its managed GCP counterpart, Cloud Composer) for orchestrating ML workflows. Airflow and Cloud Composer are general-purpose workflow tools, not ML-specific. Cloud AI Platform Pipelines narrows the focus: it is designed specifically for ML pipelines and abstracts away cluster management entirely, removing a significant operational burden from data engineering and MLOps teams.

The Broader MLOps Ecosystem Context

The launch touches on a broader competitive dynamic in the ML infrastructure space. As one commenter noted in a Hacker News discussion on the announcement:

The battle of ML pipeline ecosystems is the engine — not so much the API. It had been Beam vs [Apache] Spark. Now, Google are changing tack and saying it is TensorFlow on Kubernetes with distributed processing vs Spark-based ML pipelines.

This framing is useful for teams evaluating ML platforms. Cloud AI Platform Pipelines represents Google’s bet on a Kubernetes-native, TensorFlow-centric execution model, in contrast to Spark-based ML pipeline approaches favored in other ecosystems (such as those built on Databricks or Amazon EMR).

What This Means for MLOps Engineers

• Reduced infrastructure overhead: Google manages the GKE cluster, so teams can focus on pipeline logic rather than cluster operations.
• SDK choice matters short-term: The TFX SDK is higher-level and more opinionated; the KFP SDK offers more flexibility. Google’s plan to merge them means teams should monitor SDK deprecation announcements.
• Kubeflow familiarity is transferable: Teams already working with open-source Kubeflow will find the concepts familiar; the managed offering primarily removes cluster setup and maintenance steps.
• Lineage and reproducibility are first-class: The MLMD integration makes this a stronger choice than generic workflow tools for teams that need audit trails for model versions and training data.
• GCP lock-in considerations: Deep integration with BigQuery and Dataflow increases productivity on GCP but should be weighed against multi-cloud or portability requirements.

Summary

Cloud AI Platform Pipelines (announced in beta in early 2020) brought a fully managed, ML-specific pipeline platform to GCP, combining the abstraction of TFX with the Kubernetes-native orchestration of Kubeflow Pipelines. For teams already invested in the GCP and TensorFlow ecosystem, it offered a significant reduction in operational complexity while providing enterprise-grade lineage tracking, reproducibility, and monitoring out of the box.

What Is Google Cloud AI Platform?

Google Cloud AI Platform is an end-to-end machine learning platform-as-a-service (ML PaaS) designed for data scientists, ML developers, and AI engineers. It provides managed services covering the full ML lifecycle — from data preparation and model training through to model serving and monitoring — giving teams a unified environment to build and deploy production-grade machine learning models on Google Cloud.

In late 2019, Google announced a set of updates that made training and deploying ML models on the platform significantly more flexible and powerful. The key areas of improvement were model development (training) and model deployment (inference).

Model Training Improvements

Custom Container Support (GA)

One of the most significant announcements was the general availability of custom container support for training on Cloud AI Platform. Previously, users were constrained to the ML frameworks and library versions that Google pre-installed on the platform. Custom container support removes that restriction entirely.

With this feature, teams can:

• Package any ML framework, algorithm, or custom library into a Docker image and run it on AI Platform — regardless of whether Google natively supports it.
• Pin exact versions of Python, CUDA, framework dependencies, and internal tooling inside the container, ensuring reproducibility across local and cloud environments.
• Build and test container images locally before pushing them to the cloud, enabling faster iteration cycles.
• Integrate AI Platform training jobs into existing CI/CD pipelines, treating model training as a first-class automated build step.

For DevOps teams managing ML infrastructure, this effectively means AI Platform training can be treated like any other containerised workload — version-controlled, tested, and deployed through standard tooling.

Flexible Machine Types and Scale Tiers

To simplify hardware selection for training jobs, Google introduced scale tiers — predefined cluster specifications based on classes of Google Compute Engine (GCE) VMs. Each scale tier is mapped to workload characteristics, so teams can select an appropriate tier without manually specifying individual VM parameters.

For teams with more specific requirements, a custom tier option allows explicit configuration of the machine types for each role in a distributed training cluster:

• Master node — coordinates the training job.
• Worker nodes — execute the bulk of computation.
• Parameter servers — manage shared model parameters during distributed training.

This distributed training architecture allows large datasets and complex models to be trained at scale by parallelising computation across multiple nodes. Both custom containers and the expanded machine type options for training were made generally available with this release.

Model Deployment and Inference Improvements

What Is Inference?

In ML operations, inference refers to running a fully-trained model against new input data to generate predictions. The Cloud AI Platform Prediction service hosts trained models and handles the infrastructure required to serve predictions at scale — including provisioning, autoscaling, and load balancing — without requiring manual cluster management.

Expanded Machine Type Selection for Prediction

Previously, the Online Prediction service offered limited compute options, supporting only one or four vCPU machine configurations. With the updated release, customers can select from a broader range of GCE machine types when hosting prediction endpoints, and can attach accelerators to reduce inference latency:

• NVIDIA T4 GPUs — suited for accelerating deep learning inference workloads.
• TPUs (Tensor Processing Units) — Google’s custom ML accelerators, designed for high-throughput TensorFlow workloads.

The service continues to handle provisioning, scaling, and serving as a managed offering, meaning engineering teams do not need to manage the underlying Kubernetes infrastructure directly.

Prediction Logging to BigQuery

A notable operational addition is the ability to log prediction requests and responses directly to BigQuery. This is a practical capability for ML teams responsible for model quality in production, enabling:

• Skew detection — identifying when the distribution of live input data diverges from the training data distribution.
• Outlier analysis — flagging unusual inputs that the model may handle poorly.
• Retraining decisions — using logged data to determine when a model’s accuracy has degraded enough to warrant retraining.

This closes an important feedback loop between model serving and model improvement, supporting more mature MLOps practices.

GKE-Backed Prediction Infrastructure

The Cloud AI Platform Prediction service is powered by Google Kubernetes Engine (GKE), which provides the elastic scalability required for production prediction workloads. Teams benefit from GKE’s autoscaling and reliability characteristics without needing to manage cluster configuration themselves.

Significance for DevOps and MLOps Teams

These updates collectively represent a meaningful step towards treating ML workloads with the same operational rigour applied to standard software deployments. Key takeaways for DevOps and platform engineering teams include:

• Custom containers bring ML training into the same Docker-based workflow already used for application services, enabling consistent CI/CD integration.
• Flexible compute for both training and inference allows right-sizing of resources rather than accepting platform defaults.
• GPU and TPU support for prediction addresses latency-sensitive inference use cases without requiring self-managed accelerator infrastructure.
• BigQuery prediction logging provides the observability data needed to detect model drift and support data-driven retraining decisions.

Following the rebranding of its ML PaaS to AI Platform at Google Cloud Next, these updates reflect Google’s continued investment in making the platform competitive for production ML deployments across organisations of varying scale and maturity.