System Architecture Overview
The Neuro-Symbolic Approachβ
EduVision ITS represents a significant leap forward in educational technology by adopting a Neuro-Symbolic AI Architecture. Unlike traditional systems that rely solely on heuristic rules (Symbolic) or purely on black-box Large Language Models (Neural), EduVision synthesizes both approaches to achieve:
- Precision & Control: Symbolic logic (Knowledge Graphs, BKT) ensures the system follows pedagogical rules and tracks mastery accurately.
- Flexibility & Empathy: Neural networks (LLMs via Together AI) provide natural language understanding, dynamic explanation generation, and conversational adaptability.
The 5-Engine Coreβ
The system is composed of five distinct, decoupled engines that interact asynchronously.
graph TD
subgraph "Cognitive Core"
LE[π Learner Engine]
PE[π§ Pedagogy Engine]
KE[ποΈ Knowledge Engine]
end
subgraph "Interaction Layer"
TE[π¬ Tutor Engine]
AE[β
Assessment Engine]
end
Student((π¨βπ Student)) <--> TE
Student --> AE
TE <--> PE
PE <--> LE
PE <--> KE
AE --> LE
1. ποΈ Knowledge Engine (The "Cortex")β
- Responsibility: Stores and retrieves domain knowledge.
- Technology: PostgreSQL + pgvector (for semantic search) + Knowledge Graph (for structural relationships).
- Function: It ingests raw content (PDFs, text), chunks them into learnable units, and generates embeddings using
sentence-transformers/all-MiniLM-L6-v2. It powers the RAG (Retrieval-Augmented Generation) pipeline.
2. π§ Pedagogy Engine (The "Strategist")β
- Responsibility: Decision making.
- Technology: Reinforcement Learning / Rule-Based Heuristics (Vygotsky's ZPD).
- Function: It decides what to teach next. It balances the difficulty of the material against the student's current ability level to maintain flow. Strategies include Socratic questioning, Scaffolding, and Feynman techniques.
3. π Learner Engine (The "Hippocampus")β
- Responsibility: Long-term memory and state tracking.
- Technology: Bayesian Knowledge Tracing (BKT) + Spaced Repetition (SRS).
- Function: It maintains a probabilistic model of the student's mastery for every concept. It predicts forgetting curves and schedules reviews, persisting state in PostgreSQL.
4. π¬ Tutor Engine (The "Voice")β
- Responsibility: Natural language generation.
- Technology: Meta Llama 3.1 (8B Instruct Turbo) via Together AI API.
- Function: It translates the Pedagogy Engine's strategy into human-like dialogue. It can explain, hint, encourage, and question the student using Socratic methods. It is strictly prompted to be an educational guide, not just a chatbot.
5. β Assessment Engine (The "Judge")β
- Responsibility: Evaluation and feedback.
- Technology: AST Analysis (for Code), Semantic Similarity (for Text), LLM-based Grading.
- Function: It grades student inputs (code, essays, answers) objectively and provides structured feedback to the Learner Engine.
Data Flow Lifecycleβ
- Ingestion: Teacher uploads material -> Knowledge Engine structures it & Vectorizes it.
- Session Start: Student logs in -> Learner Engine retrieves profile -> Pedagogy Engine selects initial topic.
- Interaction: Tutor Engine presents topic -> Student responds.
- Assessment: Assessment Engine evaluates response -> Updates Learner Engine.
- Adaptation: Learner Engine updates mastery -> Pedagogy Engine adjusts strategy (e.g., move to next topic or remediate).