Žiga Vaupot's Blog

In my two previous blog posts Talking to Oracle Database in plain English and Talking to Oracle Database, this time in plain Slovenian I have been playing with Select AI in Oracle 23ai database. In these two blog posts I tested how Oracle 23ai feature called Select AI provides SQL access to generative AI using Large Language Models to generate SQL query which is then executed in database. In this blog post I am testing an option to use Select AI for Retrieval-Augmented Generation (RAG). Select AI with RAG augments natural language prompt by retrieving data (documents) from vector store (stored in Oracle 23ai ). With this additional content, hallucinations can be reduced and much more accurate answers could be retrieved. Setting it up Select AI is using Oracle 23ai AI Vector Search for similarity search using vector embeddings. To set the environment for Select AI with RAG two main tasks needs to be performed: set up vector store in Object Storage and create vector...

HROUG  hosted its 2024 's conference between16th and 19th October, as usual, in Rovinj, Croatia. This year, I was invited to present the two of my presentations: Extending Oracle Analytics Machine Learning Capabilities with OCI Services and Talking to Oracle Database in plain English … i Hrvatski! Talking to Oracle Database in plain English … i Hrvatski! My first session was about one of the coolest additions to Oracle Database 23ai, SELECT AI . SELECT AI allows users to ask the database (questions in natural language. By enabling connections with generative Large-Language Models such as ChatGPT, OpenAI Azure, Cohere and others, users can simply ask the question and can expect results, based on data stored in the database, returned narrated or simply by datasets which can be visualised and nicely presented. In this presentation, I am describing and demonstrating how to setup Oracle Database to use LLMs and  explore how well Oracle Database speaks and understand English. And si...