Text generation

Introduction to Generative AI in Snowflake

James Cha-Earley

Senior Developer Advocate, Snowflake

Managing hotel review responses

Hotel on the coast at sunset

¹ Image generated with Google Gemini 2.5 Pro

summarize()

long_review = """
This hotel was nothing short of amazing. The best views; the best location;
and the best staff. It's clean; conveniently located to other restaurants
on the strip; although we ate at the restaurants on site and each was so good
there's really no reason to leave. Ohhh the views. The beach was free of seaweed; 
amazingly clean and plenty of space; it wasn't crowded at all. 
Stay here; you will not regret it. Did I mention how comfortable the beds were. 
Yes save yourself trouble and stay here. It's really perfect.
"""


from snowflake.cortex import summarize
summarized_review = summarize(text=long_review)

Printing the summary

print(summarized_review)

The hotel exceeded expectations with stunning views, exceptional staff,
comfortable beds, and top-notch on-site dining, making it a perfect and 
highly recommended stay.

Extracting from a database

SQL query in cell1

Python code in cell2

Generating a reply

from snowflake.cortex import complete


prompt = f"Write a short response to this hotel review: {latest_review}."


# Pass the prompt
response = complete(prompt=prompt,

Generating a reply

from snowflake.cortex import complete


prompt = f"Write a short response to this hotel review: {latest_review}."


# Specify the model
response = complete(prompt=prompt, 
                    model='llama3.1-8b',

Generating a reply

from snowflake.cortex import complete


prompt = f"Write a short response to this hotel review: {latest_review}."


# Set the temperature - control predictability of output
response = complete(prompt=prompt, 
                    model='llama3.1-8b',
                    options={
                        'temperature':0.3,

Generating a reply

from snowflake.cortex import complete


prompt = f"Write a short response to this hotel review: {latest_review}."


# Limit the number of tokens with max_tokens
response = complete(prompt=prompt, 
                    model='llama3.1-8b',
                    options={
                        'temperature':0.3,
                        'max_tokens':150
                        }
                    )

Viewing the response

print(response)

Thank you so much for sharing your wonderful experience at the Hyatt Nice! 
We're thrilled to hear that you had a fantastic stay with us.

Fast, flexible, authentic

Text generation with AI_COMPLETE()

SELECT
  DESCRIPTION,
  AI_COMPLETE(

    'llama3.1-8b',

    PROMPT('Write a short response to this hotel review: {0}.', DESCRIPTION),

    {'temperature': 0.3, 'max_tokens': 150}

    ) AS completion
FROM HOTELS.REVIEWS
ORDER BY date DESC
LIMIT 1;

Output in Snowflake notebook

AI_COMPLETE() output in Snowflake notebook

AI_COMPLETE() is available out of the box

Text generation in Snowflake

-- Coding in SQL cells
AI_COMPLETE(
  'llama3.1-8b',
  PROMPT('Write a short response 
          to this hotel review: {0}.', 
          DESCRIPTION),
  {'temperature': 0.3, 
   'max_tokens': 150}
    )

Generate text for a batch of rows

# Coding in Python cells
complete(
    model='llama3.1-8b', 
    prompt = f"""Write a short response
        to this hotel review: 
        {latest_review}.""",
    options={'temperature':0.3,
             'max_tokens':150}
    )

Use to chain multi-step workflows

Let's practice!

Introduction to Generative AI in Snowflake