SageMaker Canvas ML Ops

Deploy and manage the lifecycle of machine learning models

the challenge

ML Ops engineers struggle to deploy their machine learning models beyond the platform where they built them. Currently, they must work through complicated technical processes like registering models in model registries, transferring them through shared hubs, or manually accessing underlying model files through separate development consoles.

The goal is to streamline the machine learning operations user experience so users can easily deploy their trained models as accessible services, enabling seamless sharing across teams and integration with business applications while supporting both instant predictions and large-scale batch processing through intuitive deployment workflows.

ROLE

Lead UX designer

DURATION

6 months (January 2024 - July 2024)

KEY RESPONSIBILITIES

  • Conducted heuristic evaluation of existing model capabilities to identify where ML Ops workflows fit

  • Built reusable components and patterns to scale to new ML problem types and models

  • Partnered with PM and Engineering to prioritize MLP requirements and inform future roadmap

TEAM

  • UX: 1

  • PM: 3

  • Engineering: 6 (1 SDM, 4 FEs, 1 BE)

research

During the research phase, I created a flow diagram that illustrates the end-to-end ML Ops lifecycle. This helped uncover the different surfaces where the ML Ops workflow was applicable and streamlined design ideation and execution later on.

llm and fine-tuned models

Deploy model during model evaluation

Users can deploy the LLM directly from the model catalog when reviewing its capabilities. Because foundation models are pre-trained, users can quickly productize them and evaluate performance in real time.

Users can also deploy a fine-tuned foundation model after reviewing its performance metrics, aligning with the natural progression of the ML lifecycle.

Test deployment

Users can test the performance of the deployed LLM or fine-tuned model to identify potential bias in real time and make the necessary adjustments.

time series forecasting models

Deploy model during model evaluation

Users can deploy a time series model after assessing the impact of each feature on its accuracy. This enables rapid iteration during training and continuous performance testing, allowing accuracy to improve over time.

Test deployment

Users can test the deployed time series model by selecting dataset parameters and visualizing the forecast trends. By filtering across different forecast quantiles, they can assess performance under varying scenarios and identify ways to refine the dataset for improved accuracy.

global entry points

Deploy from model versions page

In addition to being able to deploy from individual model's evaluation pages, users can also deploy a particular model version from the table without going into the model page. This accelerates the ML Ops workflow and streamlines production-level implementation.

Deploy from ML operations page

When managing the ML Ops workflow, ML ops engineers can quickly re-deploy an existing deployment or create a new deployment on the fly. This allows them to view the status of each deployment at a glance and take an action to resolve any issues during or after deployment.

View deployment details

The ML Ops engineer can view the details of a deployed model such as the deployment status, model associated with the deployment, average predictions per day, and deployment instance type. They can also copy the deployment URL to invoke the endpoint in real time, detect and resolve any potential issues.

RESULTS AND IMPACT

GA launch

GA launch

In July 2024

In July 2024

58% MoM increase

In time series models built in July 2024

58% MoM increase

In time series models built in July 2024

58% MoM increase (452 to 714)

In time series models built in July 2024

Influenced future roadmap

Built a system of patterns to scale to future ML Ops workflows and informed future roadmap and new customer requirements

Influenced future roadmap

Built a system of patterns to scale to future ML Ops workflows and informed future roadmap and new customer requirements

other projects

© 2026 Wayne Chou • UX + Visual Design. All Rights Reserved.

© 2026 Wayne Chou • UX + Visual Design. All Rights Reserved.

© 2026 Wayne Chou • UX + Visual Design. All Rights Reserved.

Create a free website with Framer, the website builder loved by startups, designers and agencies.