AI Chatbot for Proctoring

Introduction

With the rise of remote assessments, ensuring exam integrity and offering real-time support has become more essential than ever. I developed MavGuide, an AI chatbot that streamlines proctored tests by instantly addressing candidate queries, minimizing stress, and reducing operational burden for proctors.

My Role

In this project, I contributed in multiple capacities to bring MavGuide to life. I developed the underlying logic of the chatbot, crafting and rigorously testing prompts to ensure they were both accurate and user friendly. I also designed the user journey, determining the optimal placement of buttons, and creating an intuitive flow that aligned with the needs of both candidates and proctors. I also collaborated closely with the UI team to translate these designs into a visually cohesive and functional interface.

The Process

Research

I had done a competitive analysis feature comparison of various platforms that had an AI Chatbot. These include Zendesk, ProProfs Chat, Worktual, Cognigy, Indigo Chatbot, H&M Chatbot, Tidio AI, Hostinger, and Proctorio.

Based on the above research, I noted down the following recommendations for implementation:

This analysis focused on optimizing a chat system specifically for proctoring needs while learning from competitors’ best practices.

Workflow

While proctoring, candidates and proctors are the two users involved and for each user, I’ve created a workflow diagram to understand how candidates and proctors would use the chat feature with the intervention of a chatbot.

Lets start with the candidate flow:

The flowchart illustrates the interaction flow between a candidate and a chatbot proctor system. The summary of the process:

  1. The flow begins when a candidate writes a message to a proctor, with a negative count initialized to 0.
  2. After the chatbot’s introduction, there are several possible paths based on the candidate’s response:
  1. There are two ways the session can end:

Also, if the candidate is assigned to a proctor, they get 10 minutes to talk with them.

Now let’s understand the proctor flow:

    This flowchart shows the proctor’s perspective and capabilities in the chatbot system. The summary of the process:

    1. The process begins with the proctor viewing a chat listing where they can see all candidate chats.
    2. The proctor can interact with three types of situations:
    1. For each situation:
    1. Once a proctor and candidate are in conversation, the chatbot is no longer involved.
    2. At the end of the conversation, proctors have two options:

    Designs

    In the above images, the left-most panel shows the main chat dashboard where:

    The other panels show the progression of a conversation with Rick Lee:

    1. The candidate reports issues with face scan loading
    2. The system (MavGuide) provides instructions about closing other windows and handling browser alerts
    3. The candidate reports getting an error popup
    4. There’s a summary section for the proctor explaining the candidate’s face scan issue
    5. The chat shows resolution where:
    6. The chat includes options to “Take Over Chat” and “Summarize Candidate Issue”

    Conclusion

    The system is designed to handle basic inquiries via chatbot first, only escalating to a human proctor when necessary or when multiple negative interactions occur. This helps filter and manage candidate support efficiently while maintaining a clear escalation path that ensures human support is available when needed.