Snowflake allows you to store unstructured data (like PDFs, images, audio, video) in stages or using Directory Tables. You can then analyze it directly using Snowflake Cortex AI capabilities, such as Document AI, to extract information from documents using pre-trained models via SQL, while the Cortex COMPLETE Multimodal function allows direct analysis of images. You can also use Snowpark for Python, Java or Scala. Write custom code (UDFs and Stored Procedures) using libraries for OCR, NLP, image processing, etc., to process and analyze unstructured data where it resides, bringing the computation to the data.

Yes! Snowflake enables this through its AI capabilities, such as Snowflake Cortex Search, which allows you to build retrieval augmented generation (RAG) applications to find information and "chat" with your documents by asking questions in natural language. You can also use Cortex LLM Functions and build custom applications that use LLMs to understand natural language queries and extract insights or generate responses based on the content of your unstructured data (once processed or indexed).

Snowflake Document AI is specifically designed to extract data from various document formats, including PDFs (text-based and scanned), and can also handle content like handwritten text, logos and checkboxes within these documents. Snowflake Cortex's powerful multi-modal capabilities allow you to directly process and analyze image files within Snowflake. While raw audio/video analysis might require preprocessing (e.g., using Snowpark for transcription or feature extraction), the extracted text or metadata can then be analyzed using Cortex AI functions.

Yes, Snowflake allows you to control access to specific AI models available through Snowflake Cortex. This is managed via a model allowlist at the account level and through more granular role-based access control (RBAC) by granting or revoking privileges to specific model objects to defined user roles.

Customer data is not used to train, re-train or fine-tune AI models made available to other Snowflake customers. If you choose to fine-tune a model using your data within Snowflake, that fine-tuned model is available exclusively for your use and is not provided to any third-party model developer or other customers.

When you use pre-trained foundation models offered through Snowflake Cortex, multiple customers are accessing the same underlying base model version provided and managed by Snowflake (or its third-party partners). However, your specific data, prompts and outputs remain isolated to your account and are not visible to or mixed with other customers' data. Snowflake enables data privacy and security within its multi-tenant architecture.