Educators Kit

A progressive curriculum for teaching AI/ML with local models.

9 modules. From "What is a token?" to "I contributed to real AI research." Free. Local. Visual. No cloud costs. No student API keys.

Why LLMxRay for Teaching?

Advantage	Details
Free	No API keys, no cloud costs, no student subscriptions. Zero budget required.
Local	All data stays on the student's machine. No privacy concerns, no institutional data policies.
Visual	Students see tokens stream in real time, watch confidence coloring, explore reasoning chains. Abstract concepts become tangible.
Safe	No risk of students accidentally running up cloud bills or exposing sensitive data.
Cross-platform	Runs on Windows, macOS, and Linux. Deploy via npx llmxray, Docker, or git clone.
Research-ready	Every experiment is reproducible. Students contribute real findings to the open-source community.

Course Integration

LLMxRay fits naturally into:

• Introduction to AI/ML — Understanding how language models generate text
• Natural Language Processing — Token analysis, embeddings, semantic similarity
• Software Engineering — API integration, tool calling, prompt engineering
• Data Science — Benchmarking methodology, statistical analysis of model outputs
• AI Ethics — Exploring model biases, hallucination detection, language fairness

---

The Curriculum

The Journey: See → Measure → Question → Discover → Build → Contribute

Each module builds on the previous one. Students progress from observation to experimentation to original research.

Module	Title	Duration	Difficulty	The Aha Moment
1	What Is a Token?	45 min	Beginner	AI doesn't think in words — it thinks in tokens
2	How Does Temperature Work?	60 min	Beginner	Temperature isn't a linear dial — it's a phase transition
3	Can AI Lie?	90 min	Intermediate	High confidence does not mean truth
4	What Does the Model See?	45 min	Intermediate	Embeddings capture topic, not sentiment
5	When Does the Model Forget?	60 min	Intermediate	Context isn't memory — it's a sliding window
6	Can AI Use Tools?	60 min	Intermediate	Tool calling is pattern matching, not understanding
7	How Do Models Compare?	90 min	Advanced	No model is universally best
8	The Full Picture	120 min	Advanced	You can contribute to real AI research
9	What Words Cost	60 min	Intermediate	Tokenizers have a language bias — same meaning, different cost

Available Modules

All 9 modules are available now. Each module contains hands-on exercises with LLMxRay, conceptual background grounded in published research, and assessment options.

---

Quick "Aha Moment" Labs

Standalone 15-minute exercises for any lecture — no curriculum commitment needed:

Lab	The Surprise	Time
The Hallucination Lab	Models confidently fabricate history	15 min
The Tokenizer Bias	Same sentence, 5x more tokens in some languages (Module 9 covers this in depth)	15 min
The Parrot Test	Models can't repeat text verbatim — they generate, not recall	15 min
The System Prompt Leak	Students extract a "secret" system prompt	15 min
The Temperature Art Gallery	Same prompt at 8 temperatures displayed as a gallery	15 min
The Reasoning Reveal	Watch DeepSeek-R1 think step by step on a math problem	15 min

Detailed guides for these labs are coming soon.

---

Hardware Requirements

Minimum (small models only)

• RAM: 8 GB
• Storage: 10 GB free
• GPU: Not required (CPU inference works)
• Models: 1B-3B parameter models (e.g., llama3.2:1b)

Recommended (for all modules)

• RAM: 16 GB
• Storage: 20 GB free
• GPU: 6+ GB VRAM (significantly faster inference)
• Models: Up to 7B-8B parameter models

Computer Lab Setup

Option A — Per-student install: Each machine runs Ollama + LLMxRay independently.

# On each machine:
ollama pull llama3.2
npx llmxray

Option B — Shared Ollama server: One powerful machine runs Ollama, students connect via LLMxRay Settings.

# On the GPU server:
OLLAMA_HOST=0.0.0.0 ollama serve

# On each student machine:
npx llmxray --ollama-url http://gpu-server:11434

Option C — Docker deployment:

docker compose -f docker-compose.example.yml up

---

Assessment Options

Each module includes multiple assessment formats. Choose based on your course:

Format	Best for	Typical module
Written reflection (300 words)	Individual, any course level	Modules 1, 4, 5
Data analysis table + report	Data science, NLP courses	Modules 2, 3, 7
Slide deck / presentation	Group work, seminars	Modules 3, 7
Live demo + code review	Software engineering	Module 6
Full research report	Advanced / capstone	Module 8

---

Getting Help

• GitHub Discussions — Help — Ask questions about setup or usage
• Documentation — Full user guide
• Bug Reports — Report issues with structured templates

---

Using LLMxRay in your course? We'd love to hear about it. Share your experience in GitHub Discussions.