Text-to-Collection
Tool that converts natural text into structured data tables and enables large modifications through column-based —without single editing.
Background 📖
In many organizations, users already have data in textual or semi-structured formats, such as CSV, JSON snippets, system logs or AI-generated templates.
After bringing this data into a platform, users commonly face several challenges:
Editing data at scale is difficult, when users cannot experiment by adjusting individual rows
Re-importing data is required whenever the origin format changes
Role & Responsibility 🎩
UX/UI Designer, Responsibility:
Define end-to-end user journeys and interaction patterns
Define the scope of the system effect
Goal 🎯
Enable users to convert natural language input into structured data tables Focusing on context.
Allow bulk data modifications without manual, row-by-row editing.
Maintain user control through clear explanations, reversible actions, and human-in-the-loop workflows.
Constraints 💬
Need to clarify which structure could work with the Assistant service.
Data model updates are currently sent to the backend as individual API calls, leaving no intermediate decision-making or review state.
Key Design Decisions 🗝️
Decision 1 : User control and visibility
📝 Context :
When using this feature, users would shift the default view to an applied state before they decide to tackle each change.
🟨 Option :
User interacts with all actions and full review within the assistant window.
Assistant window for Prompt and full review, Action on Table
Decision 2: How to show returned-suggestion?
📝 Context :
Users want to know how the data changed, then they can decide to apply it or not.
🟨 Option :
Show before and after for each item
Show 3 highlight states of suggestion: New Added, Update, Removed
🚧 Trade-offs & Limitations
Don’t allow manual editing while AI-suggestion is pending to proceed to avoid off-tracking. After implementing an Inline-editable table, we can decide which field can work in parallel.
Use 1 color to let the user focus more on indicating AI-Suggestion. Drop out 3 colours of changing types. Instead of that providing tooltips to show more details.
Don’t show Before & After because it doesn’t support:
New creation suggestion (No previous-item to compare)
Updating in details suggestion (Not every detail can be shown on the table)
Expected Outcomes 🔮
Users can:
Successfully convert text into structured tables in a single flow
Understand the impact of transformations before applying them