TMTPOST -- Google on Friday launched BigQuery AI, a comprehensive platform that enables data scientists and analysts to build and deploy machine learning models directly within BigQuery using straightforward SQL commands. The new offering consolidates the company's built-in ML capabilities, generative AI functions, vector search, and intelligent agents into a unified ecosystem designed to accelerate the entire data-to-AI workflow.
AI Generated Image
The platform eliminates the need for data movement by bringing AI models directly to users' data through simple SQL functions. This approach allows organizations to perform tasks ranging from content generation and analysis to data enrichment without requiring expertise in specialized ML frameworks.
BigQuery AI introduces role-specific agents for data engineers, data scientists, and business users, enabling natural language interactions for building pipelines, automating workflows, and extracting insights. The platform also includes new AI and ML capabilities such as TimesFM, a pre-trained forecasting model, and expanded support for Gemini and open-source models.
The launch represents Google Cloud's effort to democratize AI access across organizations by streamlining the machine learning lifecycle and reducing technical barriers for non-technical users seeking data-driven insights.
Multimodal Data Analysis Through Generative AI
BigQuery AI integrates large language model (LLM) and embedding models directly into SQL queries through AI functions. Users can perform content generation, summarization, structured data extraction, classification, and data enrichment tasks using standard Structured Query Language (SQL) commands. The platform's managed AI functions automatically select cost- and quality-optimized models for routine tasks including filtering, rating, and classification.
Vector search capabilities extend beyond traditional keyword matching to enable semantic searches based on meaning and context. This functionality powers applications including retrieval-augmented generation, multimodal search, data deduplication, clustering, and recommendation engines. The embedding and search functions help organizations uncover conceptually related items that simple keyword searches would miss.
Complete Machine Learning Lifecycle Management
BigQuery AI handles the entire machine learning (ML) lifecycle without requiring data movement or infrastructure management. Users can train and run models directly in BigQuery using SQL or Python, managing everything from feature engineering to model training, evaluation, tuning, deployment, and inference within a single platform.
The system offers flexibility through built-in models, custom model imports, and pre-trained models like TimesFM for zero-shot inference. Unified inference capabilities support batch processing, real-time streaming, and remote execution. Users can work in their preferred environment, including BigQuery Studio, Colab Enterprise notebooks, or external IDEs.
PUMA demonstrated the platform's effectiveness by using BigQuery's integrated ML capabilities to create sophisticated audience segments based on purchase propensity, achieving a 149.8% increase in click-through rate, a 4.6% rise in conversion rate, and a 6% boost in average order value.
Role-Specific Agents and Development Tools
BigQuery AI includes three specialized agents designed for different user types. The Data Engineering Agent builds and manages data pipelines through natural language descriptions, translating requests into production-ready SQL code for data cleaning, transformations, and schema modeling. The Data Science Agent automates end-to-end workflows by creating multi-step plans, executing code, and generating visualizations from simple prompts.
The Conversational Analytics Agent enables business users to query data in natural language and receive actionable intelligence without technical expertise. The platform also provides assistive AI features including data canvas and code completion for routine tasks.
For custom applications, BigQuery offers the Conversational Analytics API for embedding natural language processing capabilities, the Agent Development Kit for building complex multi-agent systems, and the Model Context Protocol for standardizing AI model communication with databases and tools.
更多精彩内容，关注钛媒体微信号 （ID：taimeiti），或者下载钛媒体 App
