Write regulatory clinical trial reports with an internal AI platform

What was done

According to the case study published by Anthropic, a clinical study report can be up to 300 pages, and its production requires months of writing, review and approval; editors produced an average of 2.3 per year. NovoScribe, built with agent Claude Code on Amazon Bedrock and MongoDB Atlas, combines searching in a database of texts approved by professional experts and variables specific to each study to generate a compliant document, then handed over to the editors. According to a digital strategy official quoted by the source, the reporting time was reduced by 90%, and the review and approval cycles shortened. The tool was then extended to medical device verification protocols and patient documents.

Before you start

Tool cited by the source: Claude; other tools cited: Claude Code. Fill in the bracketed items: [coller], [tableaux, variables]. Strip confidential data that is not needed and check your tool’s access terms. Features and plans can change.

The method, in three steps.

1. 01

   Gather the context

   Spot the bracketed items in the prompt ([coller], [tableaux, variables]) and prepare what they ask for.

2. 02

   Adapt the prompt

   Copy the prompt below, replace the bracketed items, then paste it into Claude or Claude Code.

3. 03

   Reread before using

   Reread the result, correct it if needed and keep the final decision. Caveats are detailed below.

Your starting point.

Replace the bracketed items, then adapt the proposal to your situation. This prompt is written by the iadujour desk It is not a quotation from the source.

Here is a report section template validated by our experts: [coller], and the study data to be included: [tableaux, variables].

Writes the section according strictly to the model: same titles, same order, same formulas imposed.
- Insert only the values provided, with the source table reference.
- Do not make any clinical interpretations that are not included in the model.
- List at the end of the section missing or inconsistent data.
The text then goes to human rereading.

What deserves your attention