Comprehensive and Detailed 150 to 250 words of Explanation From [SnowPro® Specialty: Gen AI/Course Guide/topics]:
In Snowflake Document AI, running batch or programmatic inference against trained model builds utilizes the syntax ` < model_build_name > !PREDICT( < presigned_url > [, < model_version > ])`.
Key architectural components of this query include:
1. **Output Format:** The `!PREDICT` method returns an extraction payload formatted strictly as a Snowflake `VARIANT` (semi-structured data) containing a JSON hierarchy of extracted entity keys, extracted values, bounding-box spatial coordinates, and evaluation confidence scores. It does not output tabular, strongly typed structured columns directly unless explicitly parsed downstream using SQL colon operators.
2. **Stage Reference:** Stages referenced using standard named stage syntax (`@ < stage_name > `) without an external cloud storage uniform resource indicator (such as `s3://` or `azure://`) are defined and managed as Snowflake internal stages.
3. **Version Pinning:** The trailing integer argument (`9`) represents the explicit model version identifier assigned during iterative model training and evaluation inside the Document AI studio.
Thus, the query extracts semi-structured data from a document residing in an internal stage using a version-pinned Document AI model build.
References & Study Guide Topics:
* Snowflake Document Processing: Document AI `!PREDICT` Method Invocation
* Semi-Structured Data: Managing JSON Extraction Payloads and Confidence Scores
```
Submit