[FE] Pie Chart widget

[01painadam]

Goal

Implement a pie chart widget and summary text to visualise categorical data retrieved from RAG. The widget should support various categorical datasets, such as:

• Drivers of tree cover loss
• Tree cover extent – Percentage of forested vs. non-forested area
• Tree cover dynamics – Total gain, loss, and extent
• Carbon flux – Gross emissions, removals, and stocks
• Grassland breakdown – Natural, non-natural, and non-grassland areas over time
• Land cover for a given year – Breakdown of natural land categories
• Other categorical distributions – Country, region, species, habitat types, etc.

Designs

Example:

Chart

The pie chart provides an intuitive visualisation of categorical data distributions across the portfolio of KBAs.

Chart Features

• Coloured segments representing category proportions
• Legend for category labels
• Percentage labels on each segment
• Tooltips displaying category names and values

Title and Summary Text

Each chart is accompanied by dynamically generated summary text, tailored to the user persona via LLM processing.

Summary Content Requirements

• Quantitative data retrieved from RAG
• Qualitative descriptive data from KBA factsheets (also in RAG)
• Data sources, citations, or disclaimers for transparency (e.g. limitations, cautions)
• Title automatically defined by LLM

Edit/Save Interface

The widget allows users to refine the summary text using:

• Quick-select buttons:

  • Concise – Shortens the summary
  • Simplify – Adjusts language for accessibility
  • Explain – Adds detail and context

• Chat-based interaction: Users can edit text manually, request alternative phrasing, or modify data visualisation aspects before saving.

• "Add to Dashboard" button: Saves the final edited version to the dashboard.

Scope

In Scope

• Pie chart visualisation for categorical data
• Persona-based summary text generation using relevant RAG data for chart and text
• Chat-based interactions for summary editing
• Context-aware buttons for modifying summaries
• Save the final output to the dashboard (and in local storage)

Out of Scope

• Editing does not persist previous versions of summaries or charts
• No full customisation of chart styling beyond standard controls
• Share/export e.g. data download, share url

Job Stories & Acceptance Criteria

Chart Display

When I am viewing the dashboard,
I want to see a pie chart visualising categorical KBA data,
So that I can quickly understand the proportional distribution of key environmental indicators.

Acceptance Criteria:

• The pie chart correctly displays the selected dataset
• The chart includes a legend, percentage labels, and tooltips
• Data updates dynamically based on the selected dataset

Summary Text Generation

When I am reviewing a data visualisation,
I want to see a natural-language summary alongside the chart,
So that I can easily interpret the key takeaways without deep analysis.

Acceptance Criteria:

• The LLM generates summary text based on the selected dataset and user persona
• The summary includes quantitative data, qualitative insights, and citations where relevant

Summary Editing & Refinement

When I want to adjust the generated summary,
I want to modify the text using quick actions or chat,
So that I can tailor the insights to my communication needs.

Acceptance Criteria:

• Quick-select buttons adjust summary length and complexity
• Users can manually edit the summary or request changes via chat
• The final edited version is saved when added to the dashboard

Saving to Dashboard

When I finalise a chart and its summary,
I want to save it to my dashboard,
So that I can reference it later or share it with others.

Acceptance Criteria:

• Clicking "Add to Dashboard" saves the final edited version
• The saved version retains all edits and selections

Risks & Questions

Data Integrity

• How do we handle missing or incomplete categorical data in the RAG source?
• Should we offer fallback descriptions when certain data points are unavailable?

LLM Processing

• Should summary generation be fully automated, or should users confirm the text before saving?
• How do we ensure the LLM maintains factual accuracy and does not over-interpret distributions?

Solution Hints

• Use a structured layout with tooltips and labels for clarity
• Ensure summary text is concise but informative, adapting to user personas
• Allow simple, non-destructive edits with preview before saving

Future Work

...