Microsoft Copilot Studio: Tutorial for Beginners

Getting Started with Co-Pilot Studio

Introduction to Co-Pilot Studio

• To begin using Co-Pilot Studio, visit co-pilotstudio.microsoft.com and sign up with a work or school email for a free trial.

• The tutorial focuses on creating a customer-facing co-pilot that answers questions about the Australian tourism industry, utilizing the Tourism Australia website as a resource.

Setting Up Your Co-Pilot

• Users can create their co-pilots without needing to start from a website; various options are available including crafting topics and setting up generative actions.

• The setup process is quick, taking only a couple of minutes, during which the co-pilot is provisioned in the background.

Features of Customer-Facing Co-Pilots

• This example demonstrates building a public-facing co-pilot that does not require authentication but can also integrate internal documents like SharePoint repositories.

• The virtual assistant created will use generative answers to provide information based on user queries related to Sydney's attractions.

Understanding Generative AI in Co-Pilots

Contextual Awareness and Interaction

• The system generates responses by referencing multiple levels of information from the provided website, enhancing user interaction with contextual follow-up questions.

• Unlike traditional chatbots, this co-pilot maintains conversation context and utilizes large language models for dynamic interactions.

Limitations and Control Over Responses

• While generative AI offers extensive capabilities, there are instances where precise control over responses is necessary, especially regarding official information like visa requirements.

Conversational Orchestration in Chatbots

Understanding Conversational Paths

• The concept of conversational orchestration is crucial for chatbots, determining how they navigate conversations and respond to various inquiries.

• Effective planning is essential when building a chatbot co-pilot; it involves organizing information and defining the flow of conversation topics.

• Unlike generative models like ChatGPT, this chatbot relies solely on predefined data sources, limiting its responses to what has been programmed into it.

Creating Topics for Better Control

• When creating a topic (e.g., Visa), users can specify the conversational path using branching logic and conditions to guide interactions effectively.

• In this example, the bot redirects users to an official government site for visa information instead of generating AI responses, emphasizing control over content delivery.

Utilizing Trigger Phrases

• The system generates trigger phrases, which are variations of user inquiries that help direct the conversation towards specific topics.

• Users can edit these trigger phrases if necessary, allowing customization based on how well the bot understands user intent.

Enhancing User Experience with Visual Elements

• Users can test changes without saving them first; however, saving is recommended as good practice for maintaining updates.

• The interface allows adding multimedia elements such as images and videos to enhance user engagement within the conversation.

Finalizing Topic Design

• As users build out their topic (e.g., Visa information), real-time previews help visualize how content will appear during interactions.

Understanding Virtual Assistants and AI Integration

Introduction to Bruce, the Virtual Assistant

• Bruce introduces himself as a virtual assistant that utilizes AI for customer interactions, emphasizing the importance of transparency in chatbot communication.

• He discusses the significance of customizing conversation starters to enhance user engagement, moving away from generic responses.

Creating Customer-Facing Topics

• Bruce demonstrates how to create a topic regarding trip duration in Australia, highlighting the need for relevant information tailored to user inquiries.

• The process involves defining trigger phrases that prompt specific responses based on user questions about travel plans.

Utilizing Adaptive Cards and Entities

• The discussion shifts to using adaptive cards for improved visual presentation in conversations, enhancing user experience.

• Bruce explains the concept of entities within AI, clarifying their role in identifying specific categories of information from unstructured text.

Branching Conversations with Conditions

• He illustrates how to implement conditions that guide conversation flow based on user input, such as determining trip length.

• By focusing on numerical inputs (e.g., weeks), Bruce shows how the system can extract relevant data from user responses effectively.

Enhancing User Interaction through Dynamic Chaining

• The tutorial progresses into adding branching logic based on user-defined variables, allowing for more personalized interactions.

• Bruce emphasizes the importance of controlling conversation threads while still leveraging generative AI capabilities for dynamic responses.

Previewing New Features: Dynamic Chaining with Generative Actions

• A preview feature called "Dynamic chaining with generative actions" is introduced, which promises significant enhancements in conversational design and responsiveness.

Transition to Large Language Model How is the System Evolving?

Introduction of Dynamic Chaining

• The system is shifting from traditional co-pilot chatbots to a large language model, enhancing natural language understanding.

• This new approach utilizes dynamic chaining, allowing for more fluid and context-aware interactions based on user queries.

Importance of Descriptions

• Automatically generated descriptions are crucial; they help the system understand and respond accurately to user prompts.

• Users may initially be confused by the absence of trigger phrases, but this change aims to provide smarter results through improved orchestration.

Enhancements in User Interaction

• The introduction of a conversation map aids users in visualizing how discussions progress within the system.

• Generative answers serve as a starting point for inquiries, while structured data interactions allow for actions like creating or modifying database entries.

Actions and API Integration What Can the Bot Do?

New Action Features

• A new feature called "actions" enables bots to perform tasks such as retrieving information from databases via APIs.

• The platform now supports low-code plugins that simplify integration with various systems, making it accessible for non-coders.

Practical Applications

• Users can execute specific actions like getting weather forecasts without needing extensive coding knowledge or setup.

Dynamic Chaining and Weather Plugin Functionality

Overview of Dynamic Chaining

• Dynamic chaining orchestrates the flow of conversation based on user inputs, specifically location and unit preferences (Imperial or metric).

• The system identifies the user's intent to retrieve weather information for a specified location using generative AI.

Input and Output Management

• Inputs include location and units, which can be dynamically filled with optimal options. The system is designed to handle various input formats.

• Users can manually set specific values for units (e.g., Celsius instead of Fahrenheit), allowing customization of outputs.

Customizing Outputs

• Users have the option to either let AI generate messages dynamically or create specific messages manually, providing control over responses.

• There are numerous output options available (39 in total), enabling users to filter out unwanted responses before finalizing settings.

Publishing the Co-Pilot

• Before publishing, users must configure co-pilot details such as name and icon. A tip is provided for creating a custom icon via copilot.microsoft.com.

• Security settings are crucial; while authentication is typically required, this particular co-pilot will be publicly accessible without authentication.

Deployment Channels

• Various channels are available for deployment including Teams, Telegram, email, and Facebook. The demo website serves as an initial testing ground for functionality.