Research Platform2026

Agentic Classroom
Simulator

A configurable simulation platform modeling classroom environments with autonomous AI agents — instructors, students, teaching assistants — to study how information is transmitted, distorted, retained, and stabilized in agent societies.

ACL 2026Multi-Agent SystemsEducational Simulation

Research Objectives

  1. 01Simulate multi-round educational interactions with autonomous agents
  2. 02Compare teaching strategies under different classroom compositions
  3. 03Measure information degradation modes — loss, drift, distortion
  4. 04Observe emergent collective reasoning and credibility dynamics
  5. 05Produce visualizable, exportable results for academic publication

System Architecture

Three-tier design
UI Layer  (React / Next.js)
  ├─ Experiment Configurator
  ├─ Simulation Runner
  └─ Visualization Dashboard

Backend  (Serverless)
  ├─ Simulation Orchestrator
  ├─ Agent Runtime  (Gemini)
  └─ Evaluation Engine

Data Layer  (Supabase)
  ├─ Experiment Configs
  ├─ Transcripts
  ├─ Agent Memory Snapshots
  └─ Metrics & Results

Core Modules

01

Experiment Configuration

Reproducible runs across scenarios, rounds, seeds, agents, personas, teaching styles, and evaluation toggles. Immutable and versioned.

02

Agent System

Autonomous agents with role-defined responsibilities, persona-driven behavior, and private/shared memory. Environment-mediated communication.

03

Scenario Engine

Structured contexts — lecture, Q&A, exam review — defining actions, turn structure, exposure rules, and termination conditions.

04

Teaching Styles

Parameterized instructor behavior: direct instruction, Socratic questioning, retrieval practice, example-based learning, formative feedback.

05

Orchestrator

Multi-round execution with deterministic replay, batch runs, and configuration sweeps. Maintains global state and enforces scenario rules.

06

Knowledge Flow

Tracks information movement and mutation through knowledge units, transmission events, and transformations — paraphrase, omission, distortion.

07

Evaluation Engine

Retention, semantic drift, factual distortion, calibration, credibility emergence, collective reasoning gain. Per round, agent, role, experiment.

08

Visualization

Interaction timelines, conversation graphs, information flow networks, metric evolution charts, and scenario comparison panels.

Explore