Issue W4-MT-01 — Artificial Teacher Quality Evaluation

[Abdully-dev]

📋 Overview

A working model is not automatically a good teacher.

Evaluate whether Gemma actually teaches.

🎯 Objectives

Test:

Factual grounding
Retrieval relevance
Explanation quality
Socratic guidance
Weakness adaptation
Hallucination behavior
🔗 Call Chain
Test Question
↓
RAG
↓
Gemma
↓
Evaluation
🔗 Dependencies

Complete RAG + Gemma pipeline.

✅ Acceptance Criteria
Test cases are documented.
Poor responses are identified.
Prompt improvements are applied.
Retrieval failures are documented.
🌿 Branch
artificial-teacher-evaluation
🚨 Files Concerned
model/evaluation/
├── test_retrieval.py
└── test_generation.py
Issue W4-MT-02 — Complete Offline AI Validation
📋 Overview

Prove that the AI system works without internet.

This is a product requirement, not an optional test.

🎯 Objectives
Disconnect internet.
Restart application.
Test retrieval.
Test Gemma.
Test Artificial Teacher.
Verify local vector store.
🔗 Dependencies

Complete AI pipeline.

✅ Acceptance Criteria
No cloud API is required.
Retrieval works offline.
Gemma works offline.
Vector store works offline.
Artificial Teacher works offline.
🌿 Branch
complete-offline-ai-validation
🚨 Files Concerned
model/evaluation/
└── test_offline.py

backend/app/ai/
├── gemma_client.py
└── retrieval_service.py