Knowledge base (KB), RAG (retrieval-augmented generation) processing

Generating new content using oXygen’s Positron

We are currently exploring the role of local knowledge bases (KBs) in retrieval-augmented generation (RAG) AI processing. This post is the first in a series documenting our “sandbox” knowledge bases (created over a period of about 20 years) and how we’re using them in various Generative AI (GenAI) projects.

For this experiment we used our “cleaning the garage” knowledge base, which was both a standalone project and also part of our DITAinformationcenter.

Using the oXygen AI Positron Assistant to generate new content for the “garage” sample

We generally author our DITA files in oXygen, a multi-platform XML and XSLT editor from SyncRO Soft, Ltd., Romania. We’ve recently been trying out the capability of the oXygen AI Positron Assistant add-on (“POSY” for short) to generate new DITA-based content.

POSY features include:

  • Content generation: Automatically generate DITA topics and documentation drafts
  • Refactoring and summarizing: AI-driven refactoring for XML files, including generating short descriptions and alternate text for images
  • Error correction in DITA code
  • Translation of content while preserving DITA/xml markup

We experimented with the content generation feature. We prompted Positron to use the “washing the car” topic as a template and to produce a new topic, “To wash a truck.”

Here are the original “car” and generated “truck” outputs.

Original output topic "To wash the car"
Original output topic “To wash the car”
POSY-produced topic "To Wash a Truck"
POSY-produced topic “To Wash a Truck”

Thinking that our first request was a bit elementary, we asked POSY to create another topic, “To wash down the driveway,” for which there was not such an obvious template.

Here was the result.

A more challenging topic: To Wash Down the Driveway
A more challenging topic: To Wash Down the Driveway

We believe the topic generation capabilities of Positron show promise for technical writers, especially in documenting material with more strictly defined content and format (for example, API documentation).

Additional localization capabilities with Fluenta

Another oXygen add-on expands the AI-based translation capabilities of Positron.

Fluenta is a tool designed to simplify the localization of DITA projects. It parses a DITA map, resolves the references to all topics and subtopics, and prepares a unified XLIFF file that can be sent to a language service provider.

XLIFF file for the translation of a "garage" topic into Spanish
XLIFF file for the translation of a “garage” topic into Spanish