Azure-based Chatbot: Assistant for Students of a University in the USA

To help the university solve all its tasks, we decided to develop a smart chatbot based on LLM (large language model). The university collaborates with Microsoft Corporation, thus having full access to the infrastructure for creating such a tool.

We became partners with the educational institution and took on the development of the chatbot based on a set of Microsoft products: Azure and Dynamics base.

At first glance, it's a regular interactive chat: the user asks a question, and the bot responds. The main challenge faced by customers and developers of such tools is how to make the bot useful to the user, rather than just causing irritation and a desire to say "Get me an operator".

To develop a really useful product, we taught the bot to understand a variety of questions and recognize user intentions. All this is done using Microsoft Azure services.

We uploaded a list of questions and answers to Azure and then connected a separate Microsoft service - Question Answering. From that moment on, the bot was already able to respond, but only if the question matched exactly what was specified in Azure. Naturally, we needed to create many question variations so that the bot could interact with real people.

To achieve this goal, we generated a huge number of questions through Chat GPT. We uploaded them to Azure, and now our bot can provide consultations to real people who use natural language. You can ask questions in various formulations, and the bot will be able to find the right answer in its database.

A good chatbot should not only answer questions but also help the user perform a specific action. To recognize user intentions and what action they want to take, we connected Microsoft Orchestration Workflow.

Thus, our bot learned to understand what the user wants - to ask a question, log in to the system, or complain about a professor. This last intention triggers a feedback form where the student can indicate which professors they are dissatisfied with and why. The message is sent to the university administration via email.

1. SMS messages. We connected the Twilio API service for messaging.

2. A web widget on the university's website.

3. Telegram. Here we are testing the assistant's performance.

Student data and their conversations with the bot are stored within Microsoft Dynamics. In the future, we plan to launch automatic user segmentation by audiences and send mailings to these segments.

For example, if a user asks, "What documents are needed for admission?" we can include them in the prospective student segment. Later, this audience will receive separate mailings from the university tailored to their needs and goals.