Vertex AI or Agent Builder

[alpaca]

Hello,

I'm new to Google Cloud AI ecosystem and I'm confused of which tool should i use.

I need to build a chatbot prototype (web app) based on a catalogue of clothing. The chatbot has this requirements:

• i must define system instructions (rules, language, opening hours and personality)
• i need functions, so i can give a custom response with some templates
• the catalogue can be 1 or multiple unstructured documents (like PDF) in a bucket (example Google Cloud Storage)
• the AI must have access to the bucket and give information about each product inside the catalogue

Using Vertex AI i've only been able to send a single file, but i don't want to send it each time.

With Agent Builder, i've been able to create a datastore, connecting to a folder in a bucket, but i find confusing wich app type i should make and how to istruct the agent.

What would you suggest? 

Thanks in advance!

[ibaui]

Hi @alpaca,

Welcome to Google Cloud Community!

Google Cloud AI ecosystem offers many tools, and the best choice depends on your needs and expertise. For your clothing catalogue chatbot prototype, a combination of tools might be the most effective approach, leveraging the strengths of each. Avoid trying to force one single tool to do everything. Let's break down how to build this chatbot prototype using Google Cloud's services effectively.

System Design & Architecture

Document Understanding (for data extraction): Since your catalogue might be PDFs, you'll need Optical Character Recognition (OCR) and potentially PDF parsing to extract the text and structure the data. You could use Google Cloud Document AI to handle this. The output should be structured data (ideally a JSON or CSV file) in your GCS bucket. This structured data will be much easier for the AI to process than raw PDF text. This is a one-time or periodic process.

Datastore: Agent Builder's Datastore is perfect for storing this structured product information. The Datastore should be populated once with your extracted product data.

Agent Builder (Chatbot Engine): You'll use Dialogflow CX (Conversational Experience) within Agent Builder. This is a good choice because it offers excellent control over conversation flow, allows for custom responses, and integrates seamlessly with the Datastore. Choose the "Chatbot" app type in Agent Builder.

Building the Chatbot (Agent Builder)

Create a new Agent: Start a new Agent in Agent Builder, selecting the "Chatbot" app type.

Connect to the Datastore: Configure your Datastore to point to your GCS bucket containing the structured product data (JSON or CSV). Agent Builder should allow you to select the file directly from your bucket.

Design the Dialogflow CX Flow:

• Intents: Define intents that represent the user's actions, such as "Find for a product," "Get product details," or "Ask about opening hours."
• Entities: Create entities to represent key information in your catalogue, like product names, categories, sizes, colors, etc.
• Pages and Flows: Design the conversational flow using pages and transitions. This is where you'll use the Datastore to retrieve relevant product information.

Custom Response Templates: Agent Builder allows you to create response templates within your pages. You can use variables to dynamically insert product information retrieved from the Datastore.

System Instructions (Rules, Language, etc.)

• Opening Hours: Define these within the chatbot's initial greeting or introductory message.
• Personality: Determine the tone and style (formal, informal, friendly). This will influence your response templates.
• Language: Set the language for the chatbot (e.g., English, Spanish). Ensure your training data and responses reflect the chosen language.
• Rules: Establish clear guidelines on how the chatbot handles ambiguous requests, errors, or unknown products. Example: "If the product is not found, respond with: 'I'm sorry, I couldn't find a product matching your request.'"

Deployment

Once your Agent is designed and tested, deploy it. Agent Builder provides easy integration with various channels (web, mobile, etc.).

You can visit through this documentation for more details on Conversational Agents (Dialogflow CX).

I hope the above information is helpful.

[Aaryan]

Thank you! Found this quite helpful!