Open Source AI Framework

ACLAS Neuro-Edu SDK

Autonomous Cognitive Simulation Framework for AI-Powered Education

A research-grade, open-source multi-agent cognitive simulation framework that models human knowledge acquisition using on-device neural networks, vector-space semantics, and thermodynamic entropy analysis. Built with real backpropagation — no API wrappers, just first-principles AI.

GitHub Repository 🤗 Hugging Face Model

Interactive Demo

Try Neuro-Edu in Your Browser

20 autonomous AI agents, WebGL neural nebula, real-time telemetry — all running in-browser.

Features

What Makes Neuro-Edu Different

Not an API wrapper. Real neural networks. Real backpropagation. Real science.

On-Device Neural Kernel

Pure NumPy MLP with He initialization, ReLU/Sigmoid activation, and real stochastic gradient descent backpropagation. 3-layer architecture: 6→16→8→1. Zero cloud dependency.

Multi-Agent Social Learning

20+ autonomous agents exchange knowledge via real-time message bus. When an agent reaches Excited state, it broadcasts clarity signals to confused peers — modeling real classroom peer-to-peer diffusion.

Cognitive Entropy Analytics

Thermodynamic-inspired classroom tracking with Shannon entropy, MSE loss curves, GPA computation, CAS scores, retention rates, Shannon Diversity Index, and per-agent dropout risk prediction.

Knowledge Graph

D3.js-powered live concept relation map built from agent learning events. 5-dimension skill matrix tracks Logic, Math, Language, Memory, and Creative abilities per agent.

Architecture

Four-Layer Neural Architecture

Neuro-Edu's stack is designed from the ground up for transparency and extensibility:

Layer 0: WebGL / 3D Nebula

Three.js frontend rendering agent latent embeddings in real-time 3D space.

Layer 1: Neural Kernel

NumPy MLP with real backpropagation. Input [complexity, attention, skill_match, fatigue, prior_knowledge] → Dense(16)+ReLU → Dense(8)+ReLU → Dense(1)+Sigmoid → absorption probability.

Layer 2: Cognitive Profiles

5-Dim Skill Matrix + Agent DNA defining each agent's unique cognitive fingerprint across Logic, Math, Language, Memory, and Creative domains.

Layer 3-4: Core Engine + API

UltimateClassroom Bus, Social Learning Bus, Federated Training, Entropy Evaluator. Exposed via FastAPI REST: /api/teach, /api/train, /api/metrics, /api/graph, /api/reset.

The Science

Cognitive Absorption Model

Input: [complexity, attention, skill_match, fatigue, prior_knowledge]
→ Dense(16) + ReLU
→ Dense(8) + ReLU
→ Dense(1) + Sigmoid
Output: absorption_probability ∈ (0, 1)

Evaluation Metrics

GPA	= (knowledge_depth × 0.5 + attention × 0.3 + prior × 0.2) × 4.0
CAS Score	= Σ(absorption_i × attention_i) / n
Retention	= count(agents with knowledge > 0 AND mood ≠ Confused) / n
Diversity Index	= Shannon H = -Σ p(mood) × log₂(p(mood))
Dropout Risk	= (1−attention)×0.5 + fatigue×0.3 + (1−prior)×0.2

Pluggable LLM Engines

Swap the default TinyCognitionModel with Ollama or vLLM. Supports Llama-3, Qwen 2.5, DeepSeek-V3, Gemma-2, Phi-3, Mistral/Mixtral as cognitive engines within the multi-agent sandbox.

CLI & API

Command-Line Interface

Run simulations, training, and reports from the terminal:

                    # Simulate a classroom broadcast
                    python cli.py simulate --text "Introduction to Bayesian inference" --complexity 0.7 --domain math --agents 20

                    # Run federated training
                    python cli.py train --epochs 10

                    # Print full evaluation report
                    python cli.py report
                

REST API Endpoints

Method	Endpoint	Description
GET	/api/status	Live state of all 20 agents
POST	/api/teach	Broadcast instruction to all agents
POST	/api/train	Run federated training (real backprop)
GET	/api/metrics	Full evaluation: GPA, CAS, retention
GET	/api/graph	D3.js knowledge graph export
GET	/api/architecture	Neural network architecture summary
POST	/api/reset	Reset simulation environment

Source

Project Structure

Clean, well-documented codebase — 26 passing tests, comprehensive CI/CD:

                    aclas-neuro-edu/
                    ├── neuro_edu/               Source (PyPI namespace)
                    │   ├── model.py              TinyCognitionModel — NumPy MLP
                    │   ├── agent.py              NeuralStudentAgent — LLM + skill matrix
                    │   ├── engine.py             UltimateClassroom — social bus + metrics
                    │   ├── evaluator.py          GPA, CAS, Retention, Dropout Risk
                    │   ├── skills.py             5-dim SkillMatrix, 6 agent profiles
                    │   └── knowledge_graph.py    Lightweight concept graph for D3.js
                    ├── data/
                    │   └── knowledge_base.json    50-sample labeled training dataset
                    ├── web/
                    │   ├── index.html            Full dashboard: 3D + Charts + Telemetry
                    │   └── app.js                Three.js · Chart.js · D3.js frontend
                    ├── tests/                    26 passing tests (0.52s)
                    │   ├── test_model.py         9 tests — backprop correctness
                    │   ├── test_agent.py         7 tests — initialization, mood
                    │   └── test_engine.py        10 tests — classroom, training
                    ├── cli.py                    Command-line interface
                    ├── main.py                   FastAPI application entry point
                    └── requirements.txt
                

Philosophy

Why We Build Open-Source AI

"We don't wrap GPT. We model cognition from first principles."

At a time when AI in education means paying for an API wrapper, ACLAS chose a different path: build the math from scratch, publish the code, and let anyone inspect every gradient descent step. Neuro-Edu is the technical embodiment of ACLAS's institutional mission — radical transparency, access over gatekeeping, and AI built for human flourishing, not replacement.

Radical Transparency
Every weight update is observable. No black-box inference.

Access Over Gatekeeping
MIT License — any researcher, student, or engineer can run, fork, and extend.

AI for Human Flourishing
We model cognition to understand how knowledge transfers — and make it better.

Credibility Through Code
26 passing tests, real backpropagation, live WebGL visualization.

Get Started

Run It Yourself

1️⃣

Clone & Install

                            git clone https://github.com/aclascollege/neuro-edu.git

                            cd neuro-edu

                            pip install -r requirements.txt

2️⃣

Launch Dashboard

                            python main.py
                        

Open http://localhost:7860

3️⃣

Run Tests

                            pytest tests/ -v

                            26 passed in 0.52s ✓

Citation

Cite This Project

If you use Neuro-Edu in your research, please cite it using the BibTeX entry below. This helps others discover and build upon this work.

Academic Citation

BibTeX · MIT License · Zenodo DOI

Author: Atlanta College of Liberal Arts and Sciences
Year: 2026
DOI: 10.5281/zenodo.19743807
License: MIT — free for academic & commercial use
Tests: 26 passed — 0.52s

GitHub Zenodo DOI

neuro-edu.bib

@software{aclas_neuro_edu_2026,
  title     = {ACLAS Neuro-Edu: Autonomous Cognitive Simulation Framework},
  author    = {{Atlanta College of Liberal Arts and Sciences}},
  year      = {2026},
  doi       = {10.5281/zenodo.19743807},
  url       = {https://github.com/aclascollege/neuro-edu},
  note      = {Open-source AI education research framework},
  institution = {ACLAS, \url{https://aclas.college}}
}

Explore our open-source AI projects

ACLAS develops open-source, transparent AI tools for the future of education and academic integrity. Discover Aegis Graph, our academic verification prototype.

All Projects Aegis Graph

Open to Global Students

Bachelor's

Master's

Doctorate

Certificate Programs

Open to Global Students

Bachelor's

Master's

Doctorate

Certificate Programs

ACLAS Neuro-Edu SDK

Try Neuro-Edu in Your Browser

What Makes Neuro-Edu Different

On-Device Neural Kernel

Multi-Agent Social Learning

Cognitive Entropy Analytics

Knowledge Graph

Four-Layer Neural Architecture

Layer 0: WebGL / 3D Nebula

Layer 1: Neural Kernel

Layer 2: Cognitive Profiles

Layer 3-4: Core Engine + API

The Science

Cognitive Absorption Model

Evaluation Metrics

Pluggable LLM Engines

Command-Line Interface

REST API Endpoints

Project Structure

Why We Build Open-Source AI

Run It Yourself

Clone & Install

Launch Dashboard

Run Tests

Cite This Project

Academic Citation

Explore our open-source AI projects

Open to Global Students

Open to Global Students

ACLAS Neuro-Edu SDK

Try Neuro-Edu in Your Browser

What Makes Neuro-Edu Different

On-Device Neural Kernel

Multi-Agent Social Learning

Cognitive Entropy Analytics

Knowledge Graph

Four-Layer Neural Architecture

Layer 0: WebGL / 3D Nebula

Layer 1: Neural Kernel

Layer 2: Cognitive Profiles

Layer 3-4: Core Engine + API

The Science

Cognitive Absorption Model

Evaluation Metrics

Pluggable LLM Engines

Command-Line Interface

REST API Endpoints

Project Structure

Why We Build Open-Source AI

Run It Yourself

Clone & Install

Launch Dashboard

Run Tests

Cite This Project

Academic Citation

Explore our open-source AI projects

Are you sure ?