ReleaseBytes
snowflake Snowflake Blog ·

Snowflake Enhances Document Intelligence with New AI Functions

blogdatasnowflakeengineer
feature announcement

Snowflake is introducing native document intelligence capabilities, treating documents as first-class data to improve how businesses process and extract information. These new features, accessible via Snowflake Cortex AI Functions, allow for more accurate digitization and analysis of documents like invoices, contracts, and reports, benefiting a wide range of enterprise users and processes. The capabilities include parsing documents with structural preservation, extracting specific data fields, and enabling deep analytics across large document collections, all within Snowflake's secure environment.

  • →AI_PARSE_DOCUMENT enables structured document data extraction
  • →AI_EXTRACT automates data extraction with plain English descriptions
  • →AI_CLASSIFY (public preview) routes documents to appropriate AI pipelines
  • →AI_COMPLETE and AI_EMBED facilitate deep analytics and insights from document collections
  • →Document intelligence workflows integrate with Snowflake governance
Features (4) ›
  • AI_PARSE_DOCUMENT enables structured document data extraction

    The AI_PARSE_DOCUMENT function is introduced to convert documents into structured data. It supports OCR for text extraction and a LAYOUT mode that preserves document structure, including reading order, tables, and graphical elements for AI systems.

  • AI_EXTRACT automates data extraction with plain English descriptions

    AI_EXTRACT allows users to describe desired fields in plain English, returning structured JSON with confidence scores for downstream use. This function can be fine-tuned for specialized documents and workflows.

  • AI_CLASSIFY (public preview) routes documents to appropriate AI pipelines

    The AI_CLASSIFY function acts as a router, automatically directing different document types to specific AI processing pipelines, which helps manage high-volume, varied document workflows.

  • AI_COMPLETE and AI_EMBED facilitate deep analytics and insights from document collections

    AI_COMPLETE enables LLM reasoning across document collections for tasks like summarization and answering complex questions, while AI_EMBED converts content into vectors for semantic similarity clustering.

Enhancements (2) ›
  • Document intelligence workflows integrate with Snowflake governance

    Leveraging Cortex Search, the document processing runs entirely within Snowflake, meaning standard governance and access controls applied to enterprise data also govern what users can search and retrieve from documents.

  • Dynamic Tables streamline production-scale document processing orchestration

    Snowflake's Dynamic Tables simplify pipeline management by using declarative SQL statements for document data processing and refresh schedules, removing the need for complex orchestration systems.

Read the original announcement →

https://www.snowflake.com/content/snowflake-site/global/en/blog/document-intelligence-snowflake-cortex-ai-functions

Related releases