Azure AI Foundry portal or Azure Machine Learning studio: Which experience should I choose?
This article helps you understand when to use Azure AI Foundry portal versus Azure Machine Learning. While there's some overlap in the functionality in each experience, this article provides an overview of their capabilities and the development scenarios best suited for each platform.
Azure AI Foundry portal
Azure AI Foundry portal is a unified platform for developing and deploying generative AI apps and Azure AI APIs responsibly. It includes a rich set of AI capabilities, simplified user interface and code-first experiences, offering a one-stop shop to build, test, deploy, and manage intelligent solutions.
Is Azure AI Foundry portal right for you?
Azure AI Foundry portal is designed to help developers and data scientists efficiently build and deploy generative AI applications with the power of Azure's broad AI offerings.
Key capabilities of Azure AI Foundry portal
- Build together as one team. Your Azure AI Foundry portal hub provides enterprise-grade security, and a collaborative environment with shared resources and connections to pretrained models, data and compute.
- Organize your work. Your Azure AI Foundry portal project helps you save state, allowing you to iterate from first idea, to first prototype, and then first production deployment. Also easily invite others to collaborate along this journey.
- Use your preferred development platform and frameworks, including GitHub, Visual Studio Code, LangChain, Semantic Kernel, AutoGen, and more.
- Discover and benchmark from over 1,600 models.
- Provision Models-as-a-Service (MaaS) through serverless APIs and hosted fine-tuning.
- Incorporate multiple models, data sources, and modalities.
- Build Retrieval Augmented Generation (RAG) using your protected enterprise data without the need for fine-tuning.
- Orchestrate and manage prompts engineering and Large Language Model (LLM) flows.
- Design and safeguard apps and APIs with configurable filters and controls.
- Evaluate model responses with built-in and custom evaluation flows.
- Deploy AI innovations to Azure’s managed infrastructure with continuous monitoring and governance across environments.
- Continuously monitor deployed apps for safety, quality, and token consumption in production.
Azure Machine Learning studio
Azure Machine Learning studio is a managed end-to-end machine learning platform for building, fine-tuning, deploying, and operating Azure Machine Learning models, responsibly at scale.
Is Azure Machine Learning studio right for you?
Azure Machine Learning is designed for machine learning engineers and data scientists.
Key capabilities of Azure Machine Learning studio
- Build and train Azure Machine Learning model with any type of compute including Spark and GPUs for cloud-scale large AI workloads.
- Run automated Azure Machine Learning (AutoML) and drag-and-drop UI for low-code Azure Machine Learning.
- Implement end-to-end Azure Machine LearningOps and repeatable Azure Machine Learning pipelines.
- Use responsible AI dashboard for bias detection and error analysis.
- Orchestrate and manage prompt engineering and LLM flows.
- Deploy models with REST API endpoints, real-time, and batch inference.
Detailed feature comparison
The following table compares the key features of Azure AI Foundry portal and Azure Machine Learning studio:
| Category | Feature | Azure AI Foundry portal | Azure Machine Learning studio |
|---|---|---|---|
| Data storage | Storage solution | No | Yes, with cloud filesystem integration, OneLake in Fabric integration, and Azure Storage Accounts. |
| Data preparation | Data integration to storage | Yes, with blob storage, Onelake, Azure Data Lake Storage (ADLS) supported in index. | Yes, through copy and mount with Azure Storage Accounts. |
| Data wrangling | No | Yes, in code. | |
| Data labeling | No | Yes, with object identification, instance segmentation, semantic segmentation, text Named Entity Recognition (NER), integration with 3P labeling tools and services. | |
| Feature store | No | Yes | |
| Data lineage and labels | No | Yes | |
| Spark workloads | No | Yes | |
| Data orchestration workloads | No | No, although attached Spark and Azure Machine Learning pipelines are available. | |
| Model development and training | Code-first tool for data scientist. | Yes, with VS Code. | Yes, with integrated Notebooks, Jupyter, VS Code, R Studio. |
| Languages | Python only. | Python (full experience), R, Scala, Java (limited experience). | |
| Track, monitor, and evaluate experiments | Yes, but only for prompt flow runs. | Yes, for all run types. | |
| ML pipeline authoring tools | No | Yes, with the designer, visual authoring tool, and SDK/CLI/API. | |
| AutoML | No | Yes, for regression, classification, time-series forecasting, computer vision, and natural language processing (NLP). | |
| Compute targets for training | Serverless only for MaaS compute instances and serverless runtime for prompt flow. | Spark clusters, Azure Machine Learning clusters (MPI), and Azure Arc serverless. | |
| Train and fine-tune Large Language Models (LLMs) and foundation models | Limited to the model catalog. | Yes, with MPI-based distributed training and the model catalog. | |
| Assess and debug Azure Machine Learning models for fairness and explainability. | No | Yes, with the build-in Responsible AI dashboard. | |
| Generative AI/LLM | LLM catalog | Yes, through model catalog, LLMs from Azure OpenAI, Hugging Face, and Meta. | Yes, through model catalog LLMs from Azure OpenAI, Hugging Face, and Meta. |
| RAG (enterprise chat) | Yes | Yes, through prompt flow. | |
| LLM content filtering | Yes, through AI content safety. | Yes, through AI content safety. | |
| Prompt flow | Yes | Yes | |
| Leaderboard/benchmarks | Yes | No | |
| Prompt samples | Yes | No | |
| LLM workflow/LLMOps/MLOps | Playground | Yes | No |
| Experiment and test prompts | Yes, through playground, model card, and prompt flow. | Yes, through model card and prompt flow. | |
| Develop workflow | Yes, through prompt flow, integration with LangChain, and Semantic Kernel. | Yes, through prompt flow, integration with LangChain, and Semantic Kernel. | |
| Deploy workflow as endpoint | Yes, through prompt flow. | Yes, through prompt flow. | |
| Flow version control | Yes, through prompt flow. | Yes, through prompt flow. | |
| Built-in evaluation | Yes, through prompt flow. | Yes, through prompt flow. | |
| Git integration | Yes | Yes | |
| CI/CD | Yes, through code-first experiences in prompt flow, integrated with Azure DevOps and GitHub. | Yes, through code-first experiences in prompt flow, integrated with Azure DevOps and GitHub. | |
| Model registry | No | Yes, through MIFlow and registries. | |
| Organization model registry | No | Yes, through registries. | |
| Model deployment | Deployment options for real-time serving | Models as a Service (MaaS) online endpoints for MaaP catalog. | No |
| Deployment options for batch serving | No | Batch endpoints, Managed and unmanaged Azure Arc support. | |
| Enterprise security | AI Hub | Yes, manage and govern AI assets. | Yes, for both classical Azure Machine Learning and LLMs. |
| Private networking | Yes | Yes | |
| Data loss prevention | Yes | Yes | |
| Data classification | No | Yes, through Purview. |