Lab 05 · Interfaces & AI · You are Priya
Read every Teach first section before you copy SQL. Core asks which interface fits the job: browser, CLI, drivers, Snowpark, Notebooks, Streamlit, and where Cortex fits vs an external LLM.
Episode 05 · Start here
If Snowsight worksheets work, why does anyone install SnowSQL or drivers?
Same SQL engine, different doors. Browser for humans. CLI/scripts for automation. Drivers for apps. Snowpark when you want DataFrames that push work into Snowflake.
Morgan (later): don’t paste customer emails into an external LLM without a governance story.
Browser UI: worksheets, dashboards, admin. What you’ve been using.
Command-line client for scripts, PUT/GET, automation.
snowsql -c alpineJDBC/ODBC/Python… plus Spark/Kafka connectors for platforms.
DataFrame API (Python/Java/Scala) that executes lazily in Snowflake.
Interactive apps hosted inside the account (SiS) — data stays close.
Interactive cells (SQL/Python) for exploration and documented analysis.
Cortex functions (e.g. SENTIMENT, COMPLETE) run as SQL in Snowflake when enabled.
That is different from copying data to an external LLM website. Trials often lack access — know the idea for the exam.
Governed with roles, warehouses, and account controls.
SNOWFLAKE.CORTEX.SENTIMENT(...)Data leaves unless you built a careful integration. Sam cares.
Before Step 1
If I run CURRENT_USER in SnowSQL vs Snowsight, do I get a different Snowflake?
Same account engine. Interface changes how you connect — not the three layers underneath.
What this does: Prints identity and version — works from any SQL interface.
USE ROLE ROLE_CORE_LAB_ADMIN;
USE WAREHOUSE WH_CORE_XS;
USE DATABASE ALPINE_RETAIL;
USE SCHEMA LAB;
SELECT CURRENT_USER() AS login_user,
CURRENT_ROLE() AS role,
CURRENT_WAREHOUSE() AS warehouse,
CURRENT_VERSION() AS snowflake_version;Before Step 2
Should I rewrite every SELECT as Snowpark because it’s newer?
No. SQL for set-based questions. Snowpark when you need DataFrame/Python/Java pipelines that still push compute into Snowflake.
What this does: Confirms Alpine objects — same results in Snowsight or SnowSQL.
SHOW WAREHOUSES LIKE 'WH_CORE%';
SHOW DATABASES LIKE 'ALPINE%';
SELECT 'interfaces ok' AS status,
COUNT(*) AS product_rows
FROM LAB.PRODUCTS;If PRODUCTS is missing, re-run Lab 02 (or sql/02_objects.sql).
Before Step 3
If Cortex isn’t enabled, do not paste customer emails into a random website “just to try sentiment.”
Agreed. For the exam, know Cortex lives in-account. For Alpine, wait for proper enablement.
What this does: Shows the SQL shape. Keep commented unless your account has Cortex.
-- Uncomment only if Cortex is enabled:
-- SELECT SNOWFLAKE.CORTEX.SENTIMENT(
-- 'Alpine Retail had a great quarter'
-- ) AS sentiment;
-- SELECT SNOWFLAKE.CORTEX.COMPLETE(
-- 'snowflake-arctic',
-- 'Say hello to Priya in one short sentence.'
-- ) AS reply;Say: Use SQL for straightforward set-based queries. Use Snowpark when you want DataFrame/Python/Java/Scala logic that still executes inside Snowflake — not because “new is always better.”
Say: Cortex functions run in the Snowflake account under platform governance. An external LLM usually means data leaves unless you built a controlled integration. Exam: don’t equate “any AI” with Cortex.
Say: The Streamlit app is hosted in Snowflake (SiS) — close to the data — not merely a local laptop demo disconnected from the account.
Say: Yes. Different clients to the same account. Choose UI for interactive work, CLI for scripting/PUT-GET/automation.
Say: No. Notebooks are an interface / workspace. Queries still need a virtual warehouse (compute) and still read from storage.