-->

NEW EVENT: KM & AI Summit 2025, March 17 - 19 in beautiful Scottsdale, Arizona. Register Now! 

Using Generative AI for real-world KM solutions

Article Featured Image

For this application, Dremio applies generative models to its federated query, data virtualization, and data lakehouse platform to simply point at differentsources and harvest descriptions of them—similar to those contained in data catalogs. “It will scan the data and the schema, then try to infer the context from it to create a wiki entry that captures the essence of that,” Merced explained. What’ s noteworthy about this approach is that typical metadata harvesting involving machine learning and data profiling produces statistical information about data. Generative models can broaden this utility to include verbal descriptions of schema and sources, which accelerates the curation process. The domain-specific applicability of these capabilities is also significant. “I gave it a mock tax dataset and it was able to sit there and have a conversation and say, ‘This is the individual tax records; this is the business tax records,’ and give it more flavor based on what’s in there.”

Intelligent document processing

The conversational capacity of applying generative ML models to KM use cases manifest most acutely in NLQ. Mer- ced also referenced a use case in which he queried the Dremio-created wiki in natural language about taxi data, asking for the average trip distance in kilometers for a specific passenger account. Even though the data was represented as miles, the response was returned in kilometers— as Merced had specified. “I didn’t men- tion that the original data was in miles,” Merced remarked. “Through the way the schema was provided and the names of the columns, it captured the context. You can ask straight-up questions, and the queries will capture the essence of the question asked.”

GenAI applications for IDP involving robotic process automation are no less instructive. Chennupati codified the value of this technology for IDP in two parts. The first entails “understanding the structure and the layout of the document, and then understanding the content in a particular section so we derive the right information out of it.”

This use case attests to the multimodal applicability of foundation models, which supersede natural language capabilities to also include capabilities for “inferring or understanding a diagram, a picture, and image to the large language model and saying, what does this mean?” Chennupati
commented. With this approach, IDP systems enhanced by foundation models can find the required information for processing a claim or an invoice regardless of where it is in the document. A particularly exciting development pertaining to GenAI for IDP is a library of prompts that can streamline the prompting process for models to complete tasks. Blue Prism supplies such a library in which prompts for models are organized based on business domains and capabilities, like classification. According to Chennupati, “These prompts are published in our marketplace so that in other similar situations, we can also use those prompts. If prompts are not available in an organization’s prompt library or our marketplace, then at that point, a citizen developer can use our prompt designer screen to create those prompts.”

KMWorld Covers
Free
for qualified subscribers
Subscribe Now Current Issue Past Issues