BigQuery will have various AI agents that support specific roles within the company, while Looker introduces conversational analysis capabilities. The innovations should help companies work with their data more efficiently and purposefully, without the need for complex coding knowledge.
This was just announced at the Google Cloud Summit. For example, a data engineering agent will be added to BigQuery pipelines to help build data pipelines, transform data and automatically generate metadata. According to Google, this will reduce the many hours data engineers traditionally spend cleansing and validating data.
In addition, Google is introducing a data science agent in Colab notebook. This agent automates feature engineering, supports model selection and ensures faster iterations.
The third agent is the Looker conversational analytics agent, which enables users to communicate with data using natural language. This agent, developed in collaboration with DeepMind, can perform complex analyses and also explain how it reaches its conclusions. According to Google, Looker’s semantic layer improves accuracy by as much as two-thirds.
BigQuery Knowledge Engine as a foundation
To enable these AI solutions, Google is launching the BigQuery Knowledge Engine. This engine uses Gemini to analyze schema relationships, table descriptions, and query histories and generates metadata on the fly. This knowledge forms the basis for insights and semantic searches within BigQuery.
A notable point is that all Gemini-driven functions in BigQuery and Looker become available within the existing pricing models. Google emphasizes that this functionality does not require extra costs or add-ons.
Unstructured data gets a fully-fledged place
Another innovation is that unstructured data will play a more prominent role within BigQuery. With the new multimodal tables, companies can store and query complex data types such as images, audio and video alongside structured data.
The BigQuery AI Query Engine allows users to perform advanced analyses in which both structured and unstructured data are processed together with contextual information. A data analyst could ask, for example: “Which products are on these social media images?” The engine would then process the images and match them with the product catalog.
Techzine will be present at Google Cloud Next this week. Keep an eye on the website for the latest developments at the conference.
Tip: Gemini 1.0 Pro accessible via SQL in Google BigQuery and AlloyDB