ai_classify v2 + ai_extract v2.1 + sentiment via ai_classify (AI Functions demo notebooks)

[srikantdas11]

Summary

• Updates the 5 demo notebooks in 2026-05-ai-functions-data-warehouse-use-cases/ to the latest AI Functions syntax and the GA-only pre-publish gate.
• All examples now run on GA-only AI Functions.
• All ai_classify calls use v2 JSON STRING label syntax + :response[0]::STRING cast.
• All ai_extract calls use v2.1 JSON STRING schema (typed schema in Notebook 04) + :response:field:value::TYPE path access.
• ai_query calls that duplicated task-specific functionality are now ai_extract. The one remaining ai_query (generative "next_step" in Notebook 03) uses databricks-gpt-oss-120b.
• ai_analyze_sentiment (Preview) replaced with ai_classify over four sentiment labels (positive, negative, neutral, mixed) so the example is GA-runnable.
• README.md "Functions Used" column refreshed to match.
• DBC archive regenerated from the tested workspace state.

Notebook	Substance
01_document_intelligence.py	ai_query → ai_extract; VARIANT path access replaces from_json STRUCT
02_communication_triage.py	ai_classify v1 → v2 + cast
03_summarization.py	Split into ai_extract (facts) + ai_query for next_step on databricks-gpt-oss-120b
04_translation_normalization.py	ai_extract v1 → v2.1 (array form in Step 2, typed schema in Step 3)
05_sentiment_analysis.py	Sentiment via ai_classify; all ai_classify v1 → v2
README.md	Functions-used column updated

Test plan

• Run each notebook end-to-end on a Serverless SQL warehouse
• Verify the v2 VARIANT outputs match the "Expected output" tables in each notebook
• Confirm databricks-gpt-oss-120b endpoint is available in the test workspace
• Spot-check the README rendering on GitHub
• Sanity scan: no remaining claude-sonnet-4, meta-llama-3-3-70b-instruct, ai_analyze_sentiment, or v1 ARRAY(...) label form

[srikantdas11]

Made changes to the code to use the Version 2 syntax

[srikantdas11]

Code tested and Updated