Showing why minimal images matter
Minimus reduces container-image attack surface by removing unnecessary packages and vulnerabilities. Package Comparison helps users see that security advantage against the public images they already use.
The comparison showed the difference, but not the risk
The PM’s wireframe compared images by package count and vulnerability data, but didn’t show why the difference mattered. I realized the relationship between packages and attack surface needed to be immediately visible.
Make the attack surface tangible
While exploring different ways to visualize package count, the waffle chart stood out to me. It sparked the idea of representing each package as a square, while color gave me a way to layer in each package’s vulnerability severity.
Using AI to validate before polishing
I used AI tools to turn the idea into a working prototype with real data, so I could validate it with stakeholders early.
See the difference. Understand what’s driving it.
Users can compare attack surface at a glance, then inspect package count and severity to understand the difference.
In context
Mid-page on the Risk Reduction tab: after the posture chips and vulnerability comparison, above the package table it summarises.
Outcome
We don't show a demo without it. It conveys the message of the product very easily.
Sales, the graph is now in every demo
It actually shows the essence of the product.
CEO, on moving it to the homepage hero
Simplifying custom image creation
Rethinking the configuration flow
The PM’s initial wireframes put naming first, followed by a base-image dropdown and a searchable package list. Users were asked to name an image before defining its contents, with no persistent view of their selections.
I refined the sequence and selection experience to help users understand what they were building at every step.
Start with the image. Name it last.
Keep the build visible
Balance control with convenience
Package versions are grouped together, with a recommended automatic option alongside specific versions. Users can rely on Minimus to choose a compatible version or take control when needed.
Make build status visible
Creation doesn’t end when users submit an image. The gallery keeps private images and their build status in one place, making it easy to see what’s in progress, what failed, and what’s ready.
A complex build, made easier to follow
The final workflow guides developers from configuration through build, while keeping their choices and build status visible throughout.
A failed question shouldn’t end the conversation.
Shape AI lets Operations and inventory managers query business data in plain language. When a question is unclear or incomplete, a generic “try again” leaves users to figure out what went wrong. I designed recovery paths that help them move forward.
Three failures, three recovery paths
Reviewing failed conversations with the data team revealed three recurring patterns. Each called for a different response.
Give users a clear next step
I presented interpretations and related questions as selectable options within the conversation. Full questions make the next action predictable, while rephrasing remains available when the suggestions don’t fit.
Show full questions
Make each option’s meaning clear.
Keep suggestions in context
Preserve the conversational flow.
Leave room to rephrase
Let users choose another direction.
Where should users enter what’s missing?
Ask for what’s missing, where it’s needed
Based on testing, I placed the missing-input field directly within the suggested question. Users can add the required detail and continue without rewriting the question.
Raw alert tags carry too much, or too little.
BigPanda transforms monitoring tool data into alert tags. Those tags aren’t always useful as they are. Enrichment offers two ways to make them more understandable and actionable: Extraction and Composition.
The before state: a tooltip that disappeared
Help lived in a tooltip explaining both methods. It closed as soon as the user started typing, didn’t explain enough, and its examples weren’t useful.
Research: where the syntax breaks down
I set out to answer how to represent alert tags and regex syntax in a helper, and how to teach each method and its formula. Research surfaced four insights.
Composition helper: real tags, explained in place
The helper lists alert tag values that already exist in the system to pick from, with the method explained on the right. As the user types, matching tags appear. Hovering a tag shows a detailed explanation alongside.
Extraction helper: common patterns, one click away
The left side offers common patterns and syntax suggestions; the right explains the method with examples. Since most users extract values between dots or hyphens, those patterns are ready to click, and they’re done.
Five people new to the formulas, one task each
I tested the prototypes with five colleagues unfamiliar with the formulas, so they had to rely on the helper to succeed.
- All users completed the task.
- Users with little composition knowledge understood tag population after one attempt.
- The helper effectively reminded users of the syntax.
- Color made the syntax easier to understand and write.
- Explanations and examples let users with minimal regex knowledge build patterns.
- Examples were crucial to understanding the syntax.
- The color coding and split syntax backgrounds confused users.