A hands-on workshop that takes you from deep learning fundamentals to three
Physical AI applications running fully offline on an NVIDIA Jetson Nano: one
computer vision project and two language projects built on a small LLM.
The workshop runs in four parts: Foundations (sessions 1–2), Platform
(session 3), Projects (end of session 3 to session 7) and Demo day (session 8).
Who it is for. Engineering students and graduates, embedded and robotics
engineers, and software developers moving into edge AI.
Prerequisites.
Python: functions, classes, lists and dictionaries, installing packages with
pip
Linux command line: navigating folders, editing files, SSH
Math: vectors, matrices and the idea of a derivative (reviewed in session 1)
No prior machine learning experience required
✓
Session 1: Deep Learning Foundations and CNNs (4.5 hrs) Learn how neural networks learn, generalize and are measured. Then train your own image classifier and export it to ONNX.
✓
Session 2: Famous Architectures for Detection, Pose and Language (4.5 hrs) Explore YOLO, U-Net, Vision Transformers and the Transformer behind today’s LLMs. Compare models on speed and accuracy, and prompt an LLM to turn robot commands into JSON.
✓
Session 3: Jetson Nano as the Deployment Platform (4 hrs) Set up the Jetson Nano and optimize your model with TensorRT. Run your first live classifier and object detector on the edge.
✓
Session 4: Object Detection with a Safety Zone and Custom Training (4 hrs) Build an alert that triggers an LED and buzzer when a person enters a safety zone. Then train the detector to recognize your own parts. Project 1 complete.
✓
Session 5: Offline Study Assistant with RAG (4 hrs) Run a small LLM on the Nano and build a fully offline assistant that answers from the course material and cites its sources. Project 2 complete.
✓
Session 6: Voice Commander, Part 1: Robot Actions and LLM Commands (4 hrs) Turn plain sentences into validated JSON commands that safely drive a pan-tilt robot head.
✓
Session 7: Voice Commander, Part 2: Voice In, Action Out (4 hrs) Add offline speech recognition and spoken replies so you can talk to your robot and hear it answer. Project 3 complete.
✓
Session 8: Integration and Demo Day (4 hrs) Integrate your projects and present a live team demo. You leave with three working projects, a final repository and a report.