Open Clinical AI Back to Catalyst

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.

0:52 · Captioned, silent · Recorded locally Open MP4 · Full walkthrough
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

  1. Write “How many patients are there?” and browse the available data.
  2. Gemma drafts SQL; Qwen requests a correction and the model repairs it.
  3. 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.

1:10 · Captioned, silent · Recorded locally Open MP4 · Full walkthrough
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

  1. Ask for monthly CD4 counts, specifying the 2026 period and output columns.
  2. Browse available data while the question stays in place. Gemma writes SQL and Qwen reviews it.
  3. 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.

1:08 · Captioned, silent · Recorded locally Open MP4 · Full walkthrough
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

  1. Continue the monthly CD4 question with a breakdown by patient gender.
  2. Gemma generates the refinement and Qwen reviews it; no manual SQL editing is shown.
  3. 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.