While the first-generation of GenAI tooling for data engineers and AI engineers focused on construction via Python, SQL users can also participate in this critical space. The theory of unstructured document parsing/chunking/embedding transformations and how this can be used in a RAG application will be presented.

This presentation will quickly move to demonstrations of showing a simple unstructured document ETL pipeline that uses SQL to create and store vector embeddings in Apache Iceberg tables. It will then move to show how to use SQL to retrieve additional context from the vector embeddings that most closely match the user’s initial request and then augment the formal request to an LLM before returning the final response.