Simplifying ML Data Integration at Scale
When I joined the SageMaker team at AWS, enterprise customers struggled with slow and fragmented cross-account data pipelines that were difficult to configure even for experienced teams. Disconnected tools, complex configurations, and limited visibility created friction that directly slowed machine learning innovation at scale. As Lead UX Designer, I led the transformation of this experience into a unified, intuitive data integration journey, enabling data scientists and ML practitioners to access, prepare, and manage data seamlessly within their workflows so they could focus on building and deploying models rather than managing infrastructure.
Led a team of 3 designers and a researcher to deliver scalable, user-centered UX for data integration across AWS. Drove strategy and alignment with cross-functional partners to support diverse enterprise use cases.
Aug 2023 - July 2025
Launched SageMaker Unified Studio visual-ETL, data connections, and data ingestions.
AWS Cross-services
AWS Glue, Amazon EMR, Amazon Athena
$2.3B AWS segment with >100k customers
Challenge
SageMaker is AWS's fully managed machine learning platform, designed to help developers and data scientists build, train, and deploy ML models at scale. However, to fully realize this potential, users needed better support in managing their data workflows.
How might we make data integration a seamless part of the ML development process rather than a technical hurdle to overcome?
Results
Simplified Data Access & Preparation
Delivered a unified, code-optional experience for connecting, preparing, and integrating data - enabling one-click access to sources, secure credential management, and visual ETL with real-time previews.
Streamlined Monitoring & Collaboration
Built dashboards for real-time quality checks, pipeline performance, and resource usage. Enabled shared integration patterns and consistent governance to reduce setup time and improve team workflows.
Accelerated ML Development
Reduced time from data ingestion to model training with seamless handoffs and support for faster experimentation and deployment.
60%
Faster to connect new data sources
70%
Faster ML projects time-to-market
3x
Faster ML model deployment
AI visual-ETL Flow
AI-assisted drag-and-drop interface for easily building and managing data pipelines.
Data Connections & Managed Ingestion
A secure, reusable interface for managing cross-account and cross-service data integrations, with real-time, low-code solution for seamless data ingestion from diverse sources.
Guided integration builder
Near real-time data sync for low-code / no-code users with reusable connections and AWS-managed ingestions.
Workflow orchestration
Visually orchestrate, test, run, and monitor different types of data integrations, making data operations efficient and intuitive.
Consolidated Dashboard
A unified interface to monitor, manage, and debug all types of data workflows, offering real-time visibility into pipeline health, performance, and dependencies. Powered by GenAI troubleshooting.
“…Amazon SageMaker Unified Studio reduces the time-to-value for our customers' data projects by up to 40%, helping us with our mission to accelerate our digital transformation journey.”
Akihiro Suzue
Head of Solutions Sector | NTT DATA
Process
Discovery & Research: We conducted user interviews, surveys, A/B testing, and competitive analysis to uncover pain points in cloud data ingestion workflows. We identified three key user personas with overlapping access needs, all hindered by inconsistent UI handoffs.
Use case summary
Ideation & Strategy: Leveraging these insights, we restructured navigation and content around user value and feasibility. We defined UX principles—clarity, modularity, and compliance—and created north-star mockups to align stakeholders on a unified ingestion experience.
Customer journey map
Wireframing & Prototyping: We created low-fidelity wireframes to explore new ingestion flows, refining them through user feedback, then developed a high-fidelity interactive prototype for usability testing.
Wireframes
Usability Testing: We conducted usability tests with a diverse group of users to validate the design and identify areas for improvement. Based on the feedback, we made necessary adjustments to the design.
Visual-ETL builder: information architecture iterations
Option 1
Pro: Enables quick toggling between visual pipeline building and data validation, supporting faster iteration and clear separation of tasks.
Con: Separating graph and data into tabs limits in-context visibility, making complex workflows harder to manage and increasing context switching.
Option 2
Pro: Supports parallel workflows with modular panes, enabling fast iteration and real-time data validation.
Con: The layout can feel crowded and less familiar to existing AWS users, increasing the learning curve.
Option 3
Pro: Prioritizes familiarity for existing customers with consistent placement of configuration, data preview, and controller, reducing learning friction.
Con: Lacks strong, in-context data preview support, making it harder to validate configurations and transformations during setup.”
Final layout: hybrid of Options 2 & 3
Balances familiarity, efficiency, and scalability, supporting config, testing, and validation without context switching with:
Top controller bar for global actions
Left node/resource explorer for navigation and reuse
Right configuration panel consistent with Glue Studio, supporting scalability
Bottom-docked data preview for large tables
Visual Design & Style Guide: We developed a cohesive visual language, including color schemes, typography, and iconography, ensuring consistency throughout the platform. We also created a design system to maintain design consistency in future updates.
Reflections
This program reinforced the importance of designing for flexibility within complexity. Our success came not from dumbing down powerful tools, but from designing interfaces that respected user expertise while offering smarter, faster pathways.
We helped move SageMaker closer to its promise: enabling ML at scale, without barriers. If I were to do it again, I would bring data scientists into prototyping earlier to co-create the low-code patterns they needed most. I’d also engage TPMs earlier to better align across roadmap constraints and service dependencies.
The result has become more than a product improvement — it set new patterns adopted across three AWS services and continues to shape how our teams approach collaborative data tooling at scale.