A data analyst has a BigQuery table containing thousands of unstructured customer support feedback records. The analyst needs to summarize the text and classify customer sentiment directly within BigQuery using a pretrained Google large language model (LLM), such as Gemini, without moving the data outside the data warehouse. Which approach should the analyst use to perform this text analysis?

Configure a BigQuery remote connection to Vertex AI and run queries using BigQuery generative AI SQL functions to analyze the text data directly in GoogleSQL.: ### **What BigQuery AI and Remote LLM Integration Is:** **BigQuery AI functions** and **BigQuery ML** allow SQL practitioners to leverage pretrained foundation models—such as Google's **Gemini** models hosted in **Vertex AI**—directly inside their data warehouse. By setting up a **BigQuery remote connection** via the BigQuery Connection API, BigQuery securely authenticates with Vertex AI, enabling users to execute large language model (LLM) inference on tabular and unstructured text data using standard **GoogleSQL** syntax. ### **How it meets the requirements:** - **In-place text analysis (✓):** The analyst can summarize text, classify sentiment, and extract structured insights without exporting or migrating data out of BigQuery tables. - **Direct GoogleSQL interface (✓):** Generative AI functions and model endpoints integrate directly into standard `SELECT` statements, making AI capabilities accessible to SQL analysts. - **Pretrained Google LLMs (✓):** Leverages state-of-the-art models like Gemini via Vertex AI for advanced natural language understanding and generation. - **Scalable and managed (✓):** BigQuery manages query execution and batch inference, scaling compute resources seamlessly. ### **Key capabilities and benefits:** - **General-purpose and managed AI functions:** Offers full transparency over prompts and model parameters or streamlined built-in functions for common tasks like classification and rating. - **Zero data movement:** Brings the machine learning model directly to the data, enhancing governance and operational efficiency. - **Standard SQL workflow:** Enables data analysts to combine LLM inferences with standard SQL joins, aggregations, and filters in a single query. - **Unified security:** Uses Cloud IAM and BigQuery connection service accounts to manage fine-grained access to external AI services. ### **Why this is optimal:** This approach eliminates the need to build custom ETL pipelines, provision separate compute infrastructure, or write external Python scripts. It delivers immediate, secure text analysis and summarization at scale directly within the analyst's existing SQL environment.