Prompt workspace for creating, organizing, and reusing personal and public prompts within an AI prompt productivity platform.
Project
Prompt optimization platform for LLMs
Tools
Figma, Claude, Jira, Confluence
Contribution
Research, Ideation, User flows, Prototyping, Final design
My work in this iteration focused on implementing an important control layer: a public/shared prompt library that becomes the backbone for a sandbox, which pulls prompts from the library and lets users run and refine them on a single AI model at a time.
This was an iteration on a live product. The core flows for creating and editing prompts were already in place.
Prompts were still treated as isolated artifacts rather than a managed asset. Users could create, tweak, and save prompts, but had no structured way to group them, share them, or discover prompts from others as a starting point for their own work.
By introducing a public prompt marketplace where users can publish, browse, and save prompts into their own workspace, we unlocked the foundations for a managed prompt ecosystem and future playground features that reuse the same prompt assets across the web app, extension, and team accounts.
Objectives and how success will be measured
Make saving a prompt from the marketplace into a structured workspace a one-step alternative to existing ways of adding new prompts.
Make it easy to find relevant prompts in the marketplace and add them to a personal library for future use.
Make the transition from public library to personal space seamless when a user starts working with a marketplace prompt.
Keep marketplace interactions lightweight so they fit naturally into the existing prompt creation flow.
Events to track:
Save prompts from marketplace, edit and restore versions, publish/hide/delete prompts, categories, and groups.
Create prompt groups, publish categories/subcategories, report content, detect spam/duplicates, handle inappropriate or removed prompts.
Add marketplace as a first-class entry in navigation and sidebar.
What shaped the solution and what trade-offs followed.
Existing prompt management (private library, editor, history, variables) defined the data model and main workflows the marketplace had to align with.
Business requirements to plug marketplace publishing and "add from marketplace to library" into this system set the minimum surface the design needed to cover.
New marketplace flows had to be embedded into the current Prompt Manager, which limited how much navigation and information architecture could change in one iteration.
The UX needed to blur the line between editing a prompt in the marketplace and in the private library, while still making ownership and state explicit, which pushed the design toward clear status labels, badges, and history views.
Users had to see a reliable history and current status for each prompt, which drove decisions around versioning, status changes, and how destructive actions (hide/delete) are communicated in the UI.
As a result of the competitive and best-practice analysis, the following patterns were adopted:
After several iterations, the solution aligned with both user and business needs. The core flows were refined into a functional, predictable, and easy-to-use system.
I went with a familiar search-and-filter layout, with filters for model, language, and difficulty. On the main page, topics are grouped into clear categories; tapping a category drills down into a scoped view with a topic list and prompt previews. Users can also run a separate search scoped by author to quickly find prompts from specific creators.nngroup+2
The prompt action buttons mirror the personal workspace pattern, so users don't have to relearn the flow. Saving a prompt from the marketplace is streamlined to just three clicks end-to-end.
The prompt details page surfaces what the prompt does, the prompt content itself with primary actions, a worked example, and a strip of similar prompts to explore related use cases.
The flow supports smart auto-fill for title, category, subcategory, compatible models, and tags, so creators start from a sensible default rather than a blank form. They can accept, regenerate, or completely override any suggested metadata and enter their own details manually.
Designing a modular, multi-agent AI scheduling system that coordinates hotel operations and learns from historical patterns.
In 1 month of redesign, I reduced the time to complete the key workflow from X to Y minutes, based on research into users' real needs and behaviors.
Your best choice designer.
I'll help you to make a choice and ready to ansfer for any questions.