AI-generated artifacts
AI descriptions for Apache Superset dashboards
The Superset catalog is thin on the outside: a dashboard has a title, not text that says why to open it or which questions it answers. AI builds descriptions bottom-up — dataset through each chart to the dashboard — and fills the index that powers plain-language search for management and adjacent teams.
- Dataset → dashboard
- Description cascade
- Catalog index
01 · Dataset
Orders dataset
- Description: orders, revenue, conversion by channel.
- Summary: sales facts over a rolling 90 days.
02 · Charts
2 charts on the board
- Margin by line → summary per product line.
- Trend by region → summary by region and period.
03 · Dashboard
finalMargin by category
- Dashboard description from dataset + chart metadata.
Why this exists
Authors know what the board is for. The catalog does not.
The analyst remembers that "Revenue deep-dive" is about margin by category. For a director it is just a line in a list. Manual descriptions do not scale across hundreds of dashboards — metadata fields stay empty, and even good intent search has to guess.
We implement an automatic cascade: not one prompt to "describe the dashboard," but three levels. Dataset first — where the numbers come from and what they contain. Then each chart on the board. Finally text for the whole dashboard from accumulated context. Finished summaries land in the catalog index and feed AI search.
How generation works
Three levels — dataset, charts, dashboard
The dashboard description is not written in a vacuum. AI works through data sources first, then each chart, and only then assembles text for the whole board.
01 · Dataset
Dataset description and short summary
AI describes the dataset from schema, metrics, grain, and relationships to other objects. A short summary is extracted from the full text — compact metadata for the next steps and the catalog index.
02 · Charts
Each chart on the board — description and summary
For every chart, AI generates a description from what the chart shows and summaries of linked datasets. Each chart gets its own short summary: what to look at, which cuts matter, why it is on the board.
03 · Dashboard
Dashboard description from accumulated context
The final text is assembled from dataset and chart summaries — why to open the board, which questions it answers, how the pieces fit together. That is what lands in the catalog and feeds intent search.
Scope of work
What we implement
Level-by-level generation pipeline
Background jobs: dataset → charts → dashboard, passing summaries between steps and re-running when Superset objects change.
Prompts and context at each level
Separate prompt templates for dataset, chart, and dashboard synthesis — with chart metadata, dataset summaries, and your team's governance rules.
Catalog index sync
Finished descriptions and summaries land in the searchable catalog view — automatically after a pipeline run or on a schedule when Superset objects change.
Foundation for AI search
Published descriptions and summaries become part of the catalog index — often the first phase before the intent-search UI when both capabilities are in scope.
Need the system fixed fast?
Stop shipping decisions on broken numbers.
We audit what is failing, repair the foundation, and work with your team until data is reliable in day-to-day decisions.
- Founders in the implementation
- Clear priorities in week one
- Fixes shipped, not just slides
- Metric ownership made explicit
- Pipeline failures caught early
- BI logic aligned across teams
Book a focused call
Tell us where trust is breaking. We will map first fixes and ownership in one working session.