How to Use Google's PaLM 2 Bard AI to Start Your Day with a Smile

Maybe you’ve already heard about Bard, an AI experiment by Google. Bard is generative AI, it can respond to questions posed in natural language.

I love jokes, so I asked Bard to tell me a joke, and I love sarcasm, so I asked it to tell me a sarcastic joke.

Behind Bard, there is a generative AI based on Google's PaLM 2 large language models (LLMs) that you can use via the Google Vertex AI API.

What I would like is to receive, each morning, a joke in my professional email so I can start my working day with a smile.

Setting Up Joke Retrieval with Kestra

We are using Kestra to schedule a workflow that will call the Vertex AI API and then send an email.

If you’re new to Kestra, check the getting started guide.

Let’s first call the text completion API to ask Bard to tell me a joke, this can be done with the following Kestra workflow. The prompt property is the question we asked Bard.

id: bard-jokes
namespace: dev
tasks:
  - id: ask-for-jokes
    type: io.kestra.plugin.gcp.vertexai.TextCompletion
    region: us-central1
    projectId: "{{vars.projectId}}"
    serviceAccount: "{{vars.serviceAccount}}"
    prompt: Please tell me a joke

This task will output Bard's response in the form of predictions.

You can see that the joke is in the content property.

Beyond Simple Text Completion

Text completion is handy but limited. To mimic a conversational style, consider using the chat completion instead. Chat completion can contain multiple messages (the conversation) and examples to provide context to the model on how to answer the question.

id: bard-jokes
namespace: dev
tasks:
  - id: ask-for-jokes
    type: io.kestra.plugin.gcp.vertexai.ChatCompletion
    region: us-central1
    projectId: "{{vars.projectId}}"
    serviceAccount: "{{vars.serviceAccount}}"
    context: In a sarcastic tone
    messages:
    - author: user
      content: Please tell me a joke

This example is simple. For more complex use cases, you can pass a list of messages to fine-tune the response.

Emailing The Joke

To send the joke via email, we will add a MailSend task that will use the output of the ChatCompletion task as the email message text.

id: send-by-email
type: io.kestra.plugin.notifications.mail.MailSend
host: localhost
port: 3025
transportStrategy: SMTP
from: lmathieu@kestra.io
to: lmathieu@kestra.io
subject: Daily Joke
htmlTextContent: "{{outputs['ask-for-jokes'].predictions[0].candidates[0].content}}"

The expression {{outputs['ask-for-jokes'].predictions[0].candidates[0].content}} will refer to the content of the first candidate of the first prediction. Predictions are the output of the ask-for-job task.

If you want to test this example, you can use this Docker command to start a GreenMail test mail server:

docker run -t -i -p 3025:3025 -p 3110:3110 -p 3143:3143 \
        -p 3465:3465 -p 3993:3993 -p 3995:3995 -p 8888:8080 \
        greenmail/standalone:2.0.0

Ensuring Content Safety & Delivery

Google PaLM API includes safety attributes in its response to mitigate issues with improper responses. We can use these attributes to avoid sending the email if the joke seems not safe.

Here I use the If task to only send a mail if the safety score is less than 0.5.

- id: safety-belt
  type: io.kestra.core.tasks.flows.If
  condition: "{{outputs['ask-for-jokes'].predictions[0].safetyAttributes[0].scores[0] < 5}}"
  then:
  - id: send-by-email
    type: io.kestra.plugin.notifications.mail.MailSend
    [...]

Last but not least, to have it executed each day at 9 AM, we can add a Schedule trigger to the flow. It uses a CRON expression to define the schedule.

triggers:
  - id: daily
    type: io.kestra.core.models.triggers.types.Schedule
    cron: "0 9 * * *"

Full Workflow

The full YAML source of the workflow looks as follows, before executing it, you should set the two variables to your GCP project identifier and service account.

id: bard-jokes
namespace: dev
variables:
  projectId: <your-project-id>
  serviceAccount: <your-sa>
triggers:
  - id: daily
    type: io.kestra.core.models.triggers.types.Schedule
    cron: "0 9 * * *"
tasks:
  - id: ask-for-jokes
    type: io.kestra.plugin.gcp.vertexai.ChatCompletion
    region: us-central1
    projectId: "{{vars.projectId}}"
    serviceAccount: "{{vars.serviceAccount}}"
    context: In a sarcastic tone
    messages:
    - author: user
      content: Please tell me a joke
  - id: safety-belt
    type: io.kestra.core.tasks.flows.If
    condition: "{{outputs['ask-for-jokes'].predictions[0].safetyAttributes[0].scores[0] < 5}}"
    then:
    - id: send-by-email
      type: io.kestra.plugin.notifications.mail.MailSend
      host: localhost
      port: 3025
      transportStrategy: SMTP
      from: lmathieu@kestra.io
      to: lmathieu@kestra.io
      subject: Daily Joke
      htmlTextContent: "{{outputs['ask-for-jokes'].predictions[0].candidates[0].content}}"

Conclusion

For more information, you can have a look at the Google Quickstarts for Generative AI and the documentation of the TestCompletion and ChatCompletion tasks.

Integrating humor with technology, as demonstrated in this guide with Bard's generative AI and Kestra, is an example of the interesting things that can be achieved. This is just a glimpse into the potential applications. Experiment with daily jokes is a fun exploration of what's possible. It's an invitation to think creatively and see how technology can be leveraged for different purposes, including brightening your day.

If you have any questions, reach out via Kestra Community Slack or open a GitHub issue.

If you like the project, give us a GitHub star and join the open-source community.