Interactive Classroom Management Simulation

Overview

Key Takeaways

• Designed an AI-driven classroom simulation for teacher training
• Integrated Storyline variables with Gemini API using JavaScript
• Applied action mapping to structure scenario-based learning
• Created constraints and feedback systems to structure learning, not mere interaction

The Problem

Early-career teachers often struggle with classroom management because they have limited opportunities to practice difficult conversations with students during training. While alternative certification programs teach management strategies, they rarely provide safe environments where teachers can practice responding to real student behavior. This particularly affects teachers' abilities to manage classroom behaviors as they relate to cultural, cognitive, and social differences as diversity in these areas is too vast for alternative certification programs to cover.

The Solution

While alternative certification programs do cover the content of modern classroom management practices including culturally responsive classroom management, they are limited in how they let teachers practice those management skills. For this reason, I created a simulation using the power of generative large language models. This allows teachers to translate the knowledge they gain on classroom management into practical skill development in a dynamic environment, better preparing them to manage the behavior of real students in real classrooms. The simulation allows teachers to practice responding to realistic student behaviors in a low-risk environment where generative AI dynamically generates student dialogue based on the teacher’s responses.

The Process

I used action mapping to develop this simulation. Action mapping is an ideal model for scenario based learning because it focuses on the actions learners must practice and take to achieve a desired goal. Because this simulation is narrowly tailored to specific classroom management strategies for early career elementary and middle school teachers, I felt its scope and modality were conducive to the action mapping model.

This action map first identifies the end goal, then the actions necessary to achieve that goal, activities that allow learners to practice those actions, and finally, the minimum information learners need to complete those activities.

Text-Based Storyboard

Once I settled on the general shape of the simulation, I created a text-based storyboard that outlined the various slides, and the four student-characters learners would interact with. Since the goal of classroom management instruction is to prepare teachers for handling diverse groups of students, I created four characters with four different external behaviors tied to four underlying causes. Jessica, who is disengaged academically and socially because of her parents' divorce, Tristan, who is caught in a self-fulfilling cycle due to issues with past teachers, Ethan, who has been diagnosed with Oppositional Defiant Disorder, and Diego, who recently immigrated to the United States and is undergoing linguistic and cultural adjustments.

Storyboard excerpt

Interactive Prototype

I created an interactive prototype using Storyline, Vyond, and Gemini's generative language API to ensure seamless functionality.

I used Vyond to create a fully animated classroom environment to immerse learners in the experience. This included transition scenes of students approaching and leaving the teacher's desk as well as looping gifs of the students moving and speaking in the classroom. Immersion via these animations was key to creating a simulation that felt as realistic as possible. This, in combination with the emotional connection facilitated by this "show don't tell" approach, helps reinforce learning.

In Storyline, I connected a series of variables and triggers to JavaScript code that called upon the Gemini generative language API. These variables allowed the conversation to flow naturally, for the characters to retain a memory of the conversation, for Gemini to provide comprehensive feedback at the end of the simulation, and for the simulation to adapt to learners based on their stated classroom management styles.

Below is a simplified excerpt showing how Storyline variables pass teacher input to the Gemini API and update the conversation state along with the full behavioral constraints for the character Jessica.

JavaScript logic used to generate Jessica's responses.

AI System Design

The most complex challenge in this simulation's development was defining behavioral constraints for each character to ensure consistent and realistic responses from the language model. I found that the best way to keep each character in character and to prevent breaks in the model was to program it as an instructional coach roleplaying as each student with a separate instructional coach observing and offering feedback.

In the above code snippet, you can see the full prompt I used to define the character Jessica. I first set up the model as an instructional coach who would not break character from Jessica, a necessary step that prevented the model from quickly derailing the conversation in illogical ways. This also made it easier to have the model consistently act within Jessica's personality parameters and keep it from telling the learner everything about Jessica's background and situation. Finally, I provided the model with the criteria for a successful interaction so that it would not solve the issues for the learner or act as an outlier case where generally applicable management strategies have no effect.

The most vital aspect of the AI system design for this roleplay training, and for trainings like it, is the introduction of pedagogical friction. Jessica and the other characters are designed to evade, deflect, and not provide immediate answers when asked general questions like "what's wrong?" This forces learners to think critically and with empathy. AI's ability to dynamically create this friction while allowing learners to come to creative and critically considered resolutions is the biggest advantage generative language AI offers in soft skills education.

Evaluation Design

This simulation is a piece of theoretical design work. If implemented, its immediate effectiveness would be measured by pre- and post- teacher confidence surveys, scenario decision accuracy metrics, and reflection journal analysis. Its long-term effectiveness would be measured by the change in escalated behavior incidents.

Limitations

In its current form, this simulation would need further refinement before deployment. It is limited by Gemini's free-tier language API, which forces some restrictions that can cause unrealistic dialogue despite current guardrails. Relying on generative language AI models, even paid models, requires stricter guardrails. Ideally, the simulation would also connect to a text-to-speech program that would simulate realistic dialogue and potentially allow the simulation to be entirely speech based.

Future Development

An expanded version of this project would also include more classroom demographic variations, a performance analytics dashboard, scenario difficulty scaling, and longitudinal teacher improvement tracking. As it is, the simulation can adapt to teachers' own descriptions of their management style. A more robust version might incorporate broader classroom management activities first, and use that data, rather than self-reported tendancies, to create simulations that are even more adaptable and more customized to individual learners.