3D Insights for Jira

Visualize Jira issues in interactive 3D charts for instant, actionable insights

Comment

Inspiration

Teams struggle to “see” patterns in large Jira issue sets. Lists and 2D charts hide relationships, bottlenecks, and outliers - especially when fields are sparse or poorly distributed.

We wanted a fast, on-platform way to explore Jira data interactively, without exporting to spreadsheets or BI tools.

LLMs can help choose meaningful visualization dimensions - but only if they are guided by real data, strict heuristics, and clear explanations. Otherwise, you get pretty but misleading charts.

What it does

3D Insights for Jira adds a 3D dashboard gadget that helps teams explore Jira issues across three meaningful dimensions.

  • Visualizes Jira issues on X/Y/Z axes directly in a dashboard gadget
  • Uses a Rovo agent to suggest optimal axes and mappings by:
    • Prioritizing numeric fields with real variance (not flat or zero-only)
    • Selecting categorical fields with diversity (not single-bucket)
    • De-prioritizing dates unless explicitly requested
  • Validates JQL, profiles field coverage and distribution, and explains:
    • why certain fields were chosen
    • why others were skipped (e.g. “no spread”, “insufficient coverage”)
  • One-click Apply suggestions in the gadget editor - no copy/paste, no context switching
  • Optional chat for narratives and alternative perspectives, with:
    • strict JSON outputs for configs
    • explanations kept outside structured data
    • safe, permission-aware Jira access

How we built it

Platform

  • Atlassian Forge app with a Jira dashboard gadget
  • Rovo agent integrated via Forge agent and action modules

Backend

  • JQL validation and field discovery via Jira REST API
  • Sampling with requested fields (*all) and defensive limits
  • Heuristics for:
    • numeric spread detection
    • categorical diversity
    • coverage thresholds (95% with a safe 90% fallback)
  • Storage-backed suggestion flow (per gadget, per axis)

Frontend

  • React-based gadget UI with Edit and View modes
  • Non-breaking Apply suggestions button that pre-fills mappings from storage
  • Optional “Open chat” entry point with compact context

Safety and correctness

  • All Jira reads respect user permissions
  • Sanitization on both backend and frontend
  • Deterministic heuristics drive suggestions
  • AI explanations are grounded in measured data, not guesses

Challenges we ran into

  • Field variability - Jira instances differ wildly and custom fields are often sparse
    • Solved by auto-discovering numeric fields and enforcing spread and diversity checks
  • Avoiding “pretty but empty” axes - 100% coverage can still be useless
    • Flat distributions are detected and skipped
  • UX without copy/paste - chat alone wasn’t enough
    • Added storage-backed Apply suggestions that works even without chat
  • Balancing guidance with control
    • Users can override suggestions; we warn but never block

Accomplishments that we’re proud of

  • Meaningful defaults - most visualizations work on the first try
  • Clear explanations - users see exactly why fields were chosen or skipped
  • Non-breaking enhancements - everything is additive and opt-in
  • True Jira-native workflow - no exports, no external tools, no context switching

What we learned

  • Coverage alone is not enough - variance and category distribution matter more for visualization
  • Explainability builds trust - users accept AI suggestions when reasons are explicit
  • Small UX touches (like Apply suggestions) dramatically improve adoption

What’s next for 3D Insights for Jira

  • Configurable thresholds (coverage, spread) and per-team presets
  • Optional advanced date binning when time axes are requested
  • Auto-detected insight patterns (clusters, bottlenecks) with drill-down links
  • Bulk apply suggestions across dashboards and filters with safe previews
  • Opt-in, privacy-first telemetry to improve recommendations
  • Support for Assets

Built With

Share this project:

Updates