Project overview

A workspace that empowers users to experiment, evaluate, and deploy AI models without deep ML expertise.

Problems I focused on

  • No way to track or compare iterations of AI experiments

  • Complex interface barriers when trying to generate AI images

  • Difficulty getting started without clear examples or templates

  • Aggressive timeline: from 0 to shipping in 3 months

My role

  • Led UX design for version control, creating a scalable pattern now adopted across the product

  • Designed the image generation playground, balancing powerful features with user simplicity

  • Developed example templates to accelerate user onboarding (later descoped due to time constraints)

Impact

  • Successfully launched Bedrock Studio within tight deadlines

  • Democratized AI development by enabling developers of all skill levels to build secure, compliant applications rapidly while adhering to responsible AI practices. Significantly reduced development time without compromising enterprise standards.

  • Bedrock Studio is a key component of what became SageMaker Unified studio

Screenshot of Amazon Bedrock Studio interface showing configuration settings for app development with Anthropic Claude v2 model. Includes sections for app name, system prompts, parameters like temperature, top P, top K, data options, guardrails, functions, and UI. A preview panel displays a mock interview conversation about UX design.

Context

Persona

  • Experimental app developer (Bedrock studio persona)

  • Production app developer (AWS persona)

Technical constraints

  • Bedrock Studio was tightly anchored to the Bedrock console, allowing us to deliver quickly.

  • Designs in many ways had a parallelism with Bedrock console patterns.

Versions

Flowchart diagram showcasing the structure of an interface with sections labeled "Chat Builder," "Flows Builder," "Image Builder," "Prompts," "Guardrails," "Functions," and "Knowledge Base." The layout includes boxes and lines indicating relationships between different components, with options like "builder," "take version," "see versions," and "view only."
Screenshot of the Amazon Bedrock Studio interface showing configurations for a model. The model selected is "Anthropic Claude v2." Parameters include temperature, Top P, and Top K with adjustable sliders. On the right, there is a text preview simulating a job interview conversation about UX design. Options to "Save" and "Create version" are visible.

Versions for builders

Customers need a way to restore or edit to previous configurations during their building process. Versions allow them to do this.

Challenges

While developing versions, prompt builder functionality was unexpectedly introduced, requiring rapid adaptation of our design system. The key challenge was creating a flexible pattern that would seamlessly work across different builder types while maintaining consistency and user experience quality. This demanded quick iteration and collaboration to ensure our patterns could scale across the entire product suite without disrupting ongoing development.

Result

Introduced a new UX pattern, later adopted across the product.

Image playground

Amazon Bedrock Studio interface with an image playground section and options for generating images such as realistic images, comic style images, and 1800s portraits.

Generate images in explore mode

Enable enterprise users to experiment with AI image models, transforming complex image generation capabilities into an accessible workspace for business use.

Challenges

  • Multi-modal support

    Designed for diverse AI models, each with unique configurations and pre-built capabilities, requiring a flexible interface that could adapt to varying model requirements.

  • Evolving platform

    Developed solutions while Bedrock console was actively expanding, demanding continuous pattern adaptation and close collaboration with parallel feature developments.

Result

  • Shipped image playground for explorers

  • Presented product rationale for building a playground mode for app developers—enabling use cases like character creation, scenario modeling, and more.

Examples

User flowchart for exploring configurations in a software environment. Text outlines steps and scenarios for choosing and testing models, displaying options for new users who seek examples, quick ways to learn, and configuration help. Screenshots depict software interface examples.

Help users get started

Enterprise users needed a faster way to start experimenting with AI models. Many struggled to write effective prompts or understand model capabilities, leading to a steep learning curve and slower adoption.

Challenges

No engineering headcount—feature was descoped before launch.

Result

Designed an examples library that would:

  • Showcase best practices for different use cases

  • Provide ready-to-use prompt templates

  • Demonstrate optimal model configurations

  • Enable quick experimentation through pre-built scenarios

Learn more about this project