Catalyst · Questions in practice
Start with a question.
See where it takes you.
Three short examples of asking, reviewing and refining questions about data. See the actual query and result, with the full prompt a click away.
Count a population · OpenELIS
How many patients are there?
96 patients. A short question becomes a query you can review and run.
Exact question, model details and transcript
What was asked
How many patients are there?
- Data
- OpenELIS demonstration data
- Query writer
- Gemma 4 12B
- Query reviewer
- Qwen 2.5 14B
- Observed review
- The reviewer requested an automatic repair; the executed model-repaired SQL returned 96. No human SQL edit.
- Recorded
- 11 September 2026, locally in light mode
Transcript
- Write “How many patients are there?” and browse the available data.
- Gemma drafts SQL; Qwen requests a correction and the model repairs it.
- Choose Get results. The recorded source returns 96 patients.
Only waiting periods are accelerated, with the speed labeled on screen. Model review is advisory; it does not establish clinical correctness.
Look at a trend · OpenMRS HIV
How many CD4 results each month?
211 CD4 results across six months. An explicit reporting question produces monthly counts.
Exact question, model details and transcript
What was asked
Count CD4 count results from 2026-01-01 through 2026-12-31 by month for a line chart. Return month as a date (the first day of each month), and result_count.
The title above summarizes the question. The recorded prompt specifies the test, date range and output columns.
- Data
- OpenMRS HIV demonstration data
- Query writer
- Gemma 4 12B
- Query reviewer
- Qwen 2.5 14B
- Observed review
- Gemma 4 12B generated the SQL; Qwen 2.5 14B approved it. No human SQL edit.
- Recorded
- 12 September 2026, locally in light mode
Transcript
- Ask for monthly CD4 counts, specifying the 2026 period and output columns.
- Browse available data while the question stays in place. Gemma writes SQL and Qwen reviews it.
- Run the query. Six monthly rows contain 211 CD4 results.
Only waiting periods are accelerated, with the speed labeled on screen. Model review is advisory; it does not establish clinical correctness.
Refine a question · OpenMRS HIV
How does that vary by gender?
The same 211 results, grouped by gender. A follow-up adds detail while preserving the monthly totals.
Exact question, model details and transcript
What was asked
Break those same CD4 results down by patient gender, including missing gender. Count each result once so the monthly totals stay the same. Keep the 2026 date range and month as a date for the line chart. Return month, gender, and result_count.
The title above summarizes the question. The recorded prompt specifies the test, date range and output columns.
This is a follow-up to the monthly CD4 question. Missing gender was requested; the recorded output contains female and male groups only.
- Data
- OpenMRS HIV demonstration data
- Query writer
- Gemma 4 12B
- Query reviewer
- Qwen 2.5 14B
- Observed review
- Gemma 4 12B generated the refinement; Qwen 2.5 14B approved it. No human SQL edit. Recorded checks confirm unchanged monthly totals.
- Recorded
- 12 September 2026, locally in light mode
Transcript
- Continue the monthly CD4 question with a breakdown by patient gender.
- Gemma generates the refinement and Qwen reviews it; no manual SQL editing is shown.
- Run and save the result. Ten grouped rows retain the same monthly totals.
Only waiting periods are accelerated, with the speed labeled on screen. Model review is advisory; it does not establish clinical correctness.