Hands-on labs that take you from a fresh NVIDIA Jetson Orin Nano to running your own LLMs, vision models, robots, and cyber‑AI agents — designed for high‑school and university students, no prior Linux experience required. Also includes setup + sample code for Raspberry Pi and regular PCs.
|
Short, click-through decks — follow along step by step:
|
The full reference — every lab in depth: CUDA, YOLO, LLMs, RAG, ROS 2, robotics, security. |
💡 New? Open the slides and follow along. Want the deep dive? The slides link straight into the matching handbook pages.
- One tool to rule the Jetson:
sjsujetsontool— containers, model serving, Jupyter, chat, and more. - Run LLMs locally:
llama.cpp(Qwen3.5, Gemma‑4) with vision, plus cloud backends (NVIDIA, OpenAI, Anthropic). - Vision & robotics: YOLO object detection, ROS 2 / Isaac ROS, LeRobot + SO‑ARM101.
- CUDA from scratch: compile and run GPU kernels inside the container.
- Cyber‑AI: vulnerability triage with tool‑calling and RAG.
Curriculum topics: Getting Started · Linux · Deep Learning & CNNs · Transformers & LLMs · RAG & Agents · Robotics (ROS 2, LeRobot) · Cyber‑AI Security.
The Jetson already has sjsujetsontool installed. Log in as student, open a terminal:
sjsujetsontool update # refresh the tool + AI container
sjsujetsontool llama # serve a local LLM (default Qwen3.5-2B) on :8080
sjsujetsontool chat # chat with it (local or cloud backends)Full walkthrough: Lab Slides ▶
git clone https://github.com/lkk688/edgeAI.git
cd edgeAI
pip install -e . # installs the edgeLLM helper packageThen from edgeLLM.utils import performance_monitor, etc.
Building the docs & slides (maintainers): see
docs/setup.md — mkdocs serve for the handbook, docs/slides/build.sh for the
Marp decks, mkdocs gh-deploy to publish.
Dr. Kaikai Liu, Ph.D. · Associate Professor, Computer Engineering · San José State University · kaikai.liu@sjsu.edu
Learn. Build. Defend. Empower with Edge AI on Jetson.