Mastering OpenAI Function Calls for Microsoft Teams Bot

Table of Contents

1. Introduction
2. Demo
3. Understanding Function Calling
4. Code Overview
5. Conclusion
6. Resources

Introduction

In this article, we will discuss a team bot that utilizes OpenAI's function calling and Microsoft Graph API. The bot serves as a personal assistant, providing information Based on the user's calendar, contact details, and tasks. We will explore the demo and Delve into the code behind it.

Demo

Let's begin by examining the demo. The team bot acts as a personal assistant, capable of retrieving information from the user's calendar, contact details, and tasks. When a question is asked, the bot utilizes the power of Microsoft Graph API and OpenAI to fetch the Relevant information and display it in a user-friendly format.

For example, if we ask the bot "Who am I?", the question is sent to the bot, which then uses Microsoft Graph API and OpenAI to obtain the desired information. Similarly, if we Inquire about our upcoming meetings or contact details, the bot retrieves the relevant data and presents it accordingly.

However, it is important to note that the scope of the bot is limited to the user's contact details, meetings, and tasks. If a question falls outside of this scope, the bot will inform the user that it is unable to assist.

Understanding Function Calling

Before we dive into the code, let's first understand what function calling is in the Context of OpenAI. Function calling allows us to extend OpenAI models by adding real-time information retrieval capabilities. For instance, if we use a chat application powered by OpenAI APIs, we can request Current weather information.

Traditionally, this would not be possible as the model's training data is limited and may not include the most up-to-date information. However, with function calling, we can inform OpenAI that our code contains functions capable of retrieving specific information, such as current weather.

When we send a question to OpenAI along with these functions, OpenAI identifies the relevant function based on the question and instructs us on which function to call and how to pass the required arguments.

Code Overview

Now that we have a basic understanding of function calling, let's examine the code for the team bot. The code is available in the team samples repository and is built using Teams Toolkit. To run the code locally, simply press F5.

The code consists of several files. The env.local file is used for local development and contains the OpenAI API Key and other required settings. The initialize file initializes the bot and handles the SSO configuration.

The user query handler file implements the Teams FX Port SSO CommandHandler interface, which handles user queries. Within this file, the handleCommandReceived method is where the main functionality lies.

First, the bot retrieves the on-behalf-of user credential and creates an authenticated Graph client. With just a few lines of code, the bot is now authorized to make requests on behalf of the logged-in user.

Next, the bot initializes OpenAI by passing the API key. The bot then sends user messages and functions to OpenAI. The user messages include a system message and user questions, while the functions describe the capabilities of the bot, such as retrieving events or tasks.

Once OpenAI responds with the function to call and its required arguments, the bot executes the corresponding function. For example, if the user asks about upcoming meetings, the bot calls the getMyEvents function, which utilizes the Graph client to fetch the events from the user's calendar.

The response from the function is then passed back to OpenAI, which converts it into natural language and sends it back to the bot for display.

Conclusion

In this article, we explored a team bot that leverages the power of OpenAI and Microsoft Graph API. The bot acts as a personal assistant, providing information based on the user's calendar, contact details, and tasks. We examined the demo and delved into the code to understand how function calling works.

With the knowledge gained from this article, developers can build their own bots with extended capabilities using function calling. By combining OpenAI and Graph API, the possibilities for creating intelligent and interactive bots are endless.

Resources

• OpenAI
• Microsoft Graph API
• Team Samples Repository