Name: Generating OAC Semantic Model Using LLMs
Start: 2025-03-25T14:30:00-0600
End: 2025-03-25T15:15:00-0600

Tuesday March 25, 2025 2:30pm - 3:15pm MDT

Room 224

Large language models (LLMs) are powerful tools for gathering data across various fields, ushering in a new era of AI-driven code generation. LLMs can generate structured data formats such as XML and SQL. By training on extensive datasets of XML documents, these models learn to understand both the syntax and semantics of XML, enabling them to produce well-formed XML documents, complete existing structures, and transform XML data into other formats.

We aim to test the hypothesis that LLMs can generate an OAC Semantic Model. The creation of a Semantic Model is time-consuming and stands to benefit significantly from improvement. We are currently investigating this approach and are eager to determine its efficacy.

Our idea is to train the LLM using XML generation techniques for the OAC Semantic Model, based on ADW Schema metadata, including table definitions, foreign keys, view definitions, table statistics, table documentation, and data catalogs. We plan to begin training with simple examples of the Semantic Model and gradually increase the complexity.

To test our hypothesis, we will use several LLMs, including GPT, Gemini, LLaMA, and Perplexity. We will evaluate multiple test cases for each model, comparing the results to see if they can effectively generate the models and whether these models can be successfully uploaded into OAC.

Speakers

Konstantin Zhernevskiy

Senior DWH Architect, Data Intensity

Konstantin is a senior DWH architect at Data Intensity with over 20 years of experience in the Oracle technology stack. For the past 10 years, he has focused on delivering data marts, data warehouses, and analytical reporting applications. His expertise includes AI, machine learning... Read More →

Tuesday March 25, 2025 2:30pm - 3:15pm MDT
Room 224

Data (not database)

Tickets

UTOUG Training Days 2025

Konstantin Zhernevskiy

UTOUG Training Days 2025

Konstantin Zhernevskiy

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!