Cortex Code CLI 工作流程示例¶

> connect to <my demo account>

> Find all tables related to customers that I have write access to

> What privileges does my role have on this database?

> Why am I getting a permissions error?

> Generate realistic looking synthetic data into <database name>. Create a table of 10000
  financial transactions where ~0.5% of them are fraudulent. Include Amount, Location,
  Merchant, and Time. Make the fraudulent ones look suspicious based on location or amount.

> Make a dummy dataset for a clinical trial of a new blood pressure medication. List 100
  patients, their age, their dosage group (Placebo vs. 10mg), and their blood pressure
  readings over 4 weeks.

> Create a customer churn dataset for a telecom company showing customer usage for 100000
  customers. Include basic demographic data such as fake names, phone numbers, US city and
  state. Also include data usage (GB), call minutes, contract length, and whether they
  cancelled their service (churn). Ensure there's a customer_id column that's unique.
  Create the data locally and then upload it to Snowflake.

> Calculate the Churn Rate grouped by state and contract length. Order the results by the
  highest churn rate first so I can see the most risky regions and contract types.

> I want to identify the heaviest data users who are also churning.

> Build an interactive Streamlit dashboard on this data with state filters and use the
  conversation so far for examples of the kinds of charts to show. Use the attached image
  as a template for visuals and branding.

> Ensure that the Streamlit app will work with Snowflake and upload it to Snowflake.
  Give me a link to access the dashboard when it's done.

> Write a Semantic View named DEMO_TELECOM_CHURN_ANALYTICS for Cortex Analyst based on
  this data. Use the semantic-view optimization skill.

> Generate a new table called customer_call_logs. Generate 50 realistic customer service
  transcripts (2-3 sentences each) as PDF files. Some should be angry complaints about
  coverage, others should be questions about billing. Then use the AI_PARSE_DOCUMENT
  function to extract the text and layout information from the PDFs into the TRANSCRIPT_TEXT
  column. Split text into chunks for better search quality.

> Create a Cortex Search Service named CALL_LOGS_SEARCH that indexes these transcripts.
  It should index the TRANSCRIPT_TEXT column and filter by CUSTOMER_ID.

> Build a Cortex Agent that has access to two tools:
  - cortex_analyst: For querying the TELECOM_CUSTOMERS SQL table.
  - cortex_search: For searching the CALL_LOGS_SEARCH service.

  Write a system prompt for this agent:
  - Persona: You are a Senior Retention Specialist.
  - Routing Logic: If the user asks for 'metrics', 'counts', or 'averages', use the
    Analyst tool. If the user asks for 'sentiment', 'reasons', or 'summaries of calls',
    use the Search tool.
  - Output Format: Always verify the customer ID before answering. If the risk score is
    high, end the response with a recommended retention offer (e.g., 'Offer 10% discount').
  - Constraint: Never reveal the raw CHURN_RISK_SCORE to the user; interpret it as 'Low',
    'Medium', or 'High'.

> Let's deploy this agent to Snowflake Intelligence.