Context

Clinical writers work with specialist terminology, existing protocols, and source material that must be understood before it is reused. Generating a fluent paragraph does not resolve those responsibilities. Gilles led product and AI technology for a clinical writing assistant proof of concept. The verified scope covered roadmapping, supporting application development, and a retrieval backend. It remained a proof of concept; this portfolio makes no claim of regulatory certification, validated clinical use, or improved patient outcomes.

Approach

Fountain-pen nib beside ink reference marks

The prototype retrieved relevant passages from clinical trial protocols using metadata filters and source-linked results. A document-editor add-in let a writer retrieve passages, generate draft text, and insert it into a working document. The general interaction principle is to keep evidence and drafting close to the writer’s existing task. Enterprise retrieval supplies related engineering context, while this example keeps its distinct experimental status and specialist review needs visible.

Quality controls

Expert review is essential when drafting can alter meaning, omit qualifications, or reuse material outside its original context. As general evaluation guidance, test source relevance, citation support, and whether generated statements preserve important constraints. Structured evaluation explains why valid fields alone cannot answer those questions. These are recommended checks, not an assertion of a completed regulatory validation programme. A passage that answers a retrieval question may still need qualification before it belongs in the document being written.

Outcomes

Oxblood document folio with ivory index tabs

The delivered proof of concept joined protocol retrieval and assisted drafting within a document workflow. It demonstrated an application path for specialist knowledge work, with human review remaining central. Further validation and any production or regulated use would require evidence beyond the prototype contribution described here.

  • Built a source-linked protocol retrieval backend.
  • Developed an editor add-in for retrieval, draft generation, and insertion.