Salvat

A knowledge guide assisted by AI to locate current content, cite its origin and know when an expert review is needed

In an editorial environment, the difficulty is not just finding information. It is to know what edition, what author, what date and what context sustain a response. When content touches health or well-being, this traceability becomes even more important.

Karmina AI Studio implemented a Knowledge Guide who exclusively consulted approved sources and responded with internal citations. The AI accelerated recovery and synthesis; the editorial team maintained authorship, judgment, and expert review.

Main CTA: Activate your editorial knowledge CTA Secondary: Discover the Knowledge Guide

Summary Fact Sheet

• Brand: Salvat
• Sector: Editorial and Health
• Challenge: to facilitate traceable internal consultations on a large fund
• Principal Agent: Knowledge Guide
• Complementary agents: Blog Publisher · Cloud Index
• Service: Knowledge Foundation
• AI layer: Knowledge infrastructure and consulting agent cited
• Human role: canonical source, scientific or editorial review, rights and publication

El challenge

A plausible answer is not an editorially valid answer

Generative tools can write fluently even if they confuse edits, mix up content, or omit a limitation. In an extensive catalog, the same question may have different answers depending on the work, the audience, and the date.

The challenge was to use the AI as a gateway to knowledge, not as a substitute for the source.

The starting point

The project began with an inventory of collections, metadata, rights, formats, dates and owners. Not all of the fund had to be available for all uses: a commercial synopsis, a protected fragment, and a health text needed different permissions and circuits.

The Cloud Index helped detect versions and gaps. Then, only approved sources went into Knowledge Guide.

The Karmina AI Studio approach

The guide should behave as a responsible consultation system:

• respond with the visible source;
• distinguish citation, synthesis and interpretation;
• indicate date and validity;
• recognize when the base did not contain the answer;
• scaling sensitive content to a specialist;
• Respect limits of rights and reproduction.

The Blog Publisher could transform a validated research into a draft, but never publish it without authorship and review.

How it worked

1. The person asked a question within their level of permission.
2. The system recovered fragments from approved sources.
3. He prepared a synthesis with title, edition, date and location.
4. He pointed out conflicts, absences or sensitive content.
5. One person validated the subsequent use.
6. The corrections updated the base, not just that response.