Skip to main content

Tutor Engine (TE)

Overview​

The Tutor Engine is the conversational interface of EduVision. It is the "Voice" that interacts directly with the student. Powered by Large Language Models (LLMs), it transforms raw pedagogical strategies into natural, empathetic, and context-aware dialogue.

Core Capabilities​

1. Neuro-Symbolic Dialogue Management​

Unlike generic chatbots, the TE follows a strict pedagogical protocol dictated by the Pedagogy Engine.

  • Strategy Injection: The prompt context includes the specific teaching strategy (e.g., "Socratic Method", "Direct Instruction", "Scaffolding").
  • Safety Rails: Prevents the LLM from giving direct answers when the goal is to guide the student.

2. Context-Aware Generation (RAG)​

To minimize hallucinations and ensure accuracy, the TE uses Retrieval-Augmented Generation (RAG).

  • Context Retrieval: Queries the Knowledge Engine for relevant excerpts.
  • Prompt Construction: Combining system instructions, conversation history, and retrieved knowledge.

3. Adaptive Tone & Style​

The TE adjusts its persona based on the student's profile (from the Learner Engine).

  • Beginner: Encouraging, simple language, more emojis.
  • Advanced: Concise, technical, challenging.

Technical Implementation​

Prompt Engineering​

The system uses a sophisticated prompt template structure:

SYSTEM_PROMPT = """
You are an expert AI Tutor.
Current Strategy: {strategy}
Student Level: {mastery_level}

CONTEXT:
{retrieved_knowledge}

INSTRUCTIONS:
- Do not reveal the answer directly if the strategy is 'Socratic'.
- Use the provided context to answer questions.
- Maintain an encouraging tone.
"""

Model Configuration​

  • Base Model: meta-llama/Meta-Llama-3-8B-Instruct (Fine-tuned on educational datasets).
  • Inference: Optimized with quantization (4-bit/8-bit) for efficient deployment.
  • Parameters: temperature=0.7 (Creative but focused), max_tokens=512.

Integration with Other Engines​

  • Pedagogy Engine: Provides the strategy and next_step instructions.
  • Knowledge Engine: Provides the factual content (context) for the response.
  • Learner Engine: Provides the student's profile and history for personalization.