SutraCore/Learn/SutraAI Jetson Lab
Free course · by the SutraCore team

Free NVIDIA Jetson AI course: 30 days, 16 steps

Orin NX 16GBJetPack 6.2.1CUDA 12.6TensorRT 10.3Python 3.10PyTorch 2.8
Jetson Orin NX 16GB · JetPack 6.2.116 steps · 30 daysDay 1 = Mon 26 Oct 2026

Build the SutraAI flow on your Jetson, one step at a time

Each page tells you where the code lives, what to run, how to test it, and when the step is done. Weeks 1–3 are Jupyter notebooks only. Web code starts after the Day 21 adoption gate.

Phase A · Days 1–21Notebooks on the Jetson: image, sensor, audio and detection models, export, profiler, device simulator, NVIDIA models
Phase B · Days 22–30ml_core package → FastAPI worker → Next.js slice
30-day plan + NVIDIA guidelines →Free courses and skills

Your Jetson

Every step is written for the board below, read from your jtop info screen. If you ever re-flash or upgrade, check these numbers again before continuing.

ItemYoursWhat it means for the steps
ModuleJetson Orin NX 16GB on the Orin dev kit carrier, Super config (P3767-0000)16 GB shared by CPU and GPU: room for an LLM next to a small training job
GPUAmpere, 1024 CUDA cores, 32 Tensor Cores, CUDA arch 8.7Build llama.cpp with CMAKE_CUDA_ARCHITECTURES=87
DLADeep-learning accelerators (the Orin NX has them; the Orin Nano doesn't)Optional extra TensorRT target in Step 05
CPU8 × Arm Cortex-A78AEUse num_workers=4 in data loaders
JetPack / L4T6.2.1 / 36.4.7 · Ubuntu 22.04 · kernel 5.15.148-tegraNo re-flash needed. Don't upgrade during the 3-week notebook phase.
CUDA · cuDNN · TensorRT12.6.68 · 9.3.0 · 10.3.0.30PyTorch from the jp6/cu126 Jetson index (Step 02)
Python3.10.12Matches the Jetson PyTorch wheels (cp310)
Python packages (installed in Step 02)torch 2.8.0 · torchvision 0.23.0 · torchaudio 2.8.0 · onnxruntime-gpu 1.23.0 · numpy 1.26.4 · matplotlib 3.10.9 · pillow 11.3.0 · scipy 1.15.3From the jp6/cu126 index, locked with constraints-jp621.txt so nothing replaces them
Model converter (Step 05)onnx2tf 2.6.9 (no TensorFlow), needs Python 3.12+Run it in Google Colab, or in a separate Python 3.12 .venv-convert made with uv
VPI · Vulkan3.2.4.0 · 1.3.204Not needed in Phase A
OpenCV4.8.0, built without CUDAFine: these steps use PIL and torchvision for images
Power mode15 W when the screenshot was takenSwitch to MAXN SUPER (sudo nvpmodel -m 0) in Step 01 before training
Monitorjtop 7.2.1 (jetson-stats)"7.2.1" is the jtop tool's version, not JetPack. Your JetPack is 6.2.1.
NetworkYour hostname and LAN IP · optional Tailscale · DockerSSH on your LAN, or over Tailscale from anywhere

How to use these pages

  1. Open the step for today. Every step shows its day and date.
  2. Each code block is labelled with where it runs:
Jetson terminalSSH session on the Jetson Notebook cellJupyterLab on the Jetson Laptop terminalyour Windows or Linux PC Filesave at the path shown
  1. Work through Do it, then run Test it. Only tick Done when items you have actually seen pass. Your ticks show up in the list below, saved in this browser only.
  2. Stuck? Every step has an If it breaks table. Courses, NVIDIA models and the glossary are on the reference page.

All steps

Week 1 · Jetson + first model

  1. 01Set up the JetsonDay 1 · Mon 26 Oct · terminal onlyNot started
  2. 02CUDA, Python and JupyterLabDay 2 · Tue 27 Oct · NB00 · NB01Not started
  3. 03Machine learning basicsDay 3 · Wed 28 Oct · NB02Not started
  4. 04Your first image modelDays 4–5 · Thu 29 – Fri 30 Oct · NB10 · NB11Not started
  5. 05Export and validateDays 6–7 · Sat 31 Oct – Sun 1 Nov · NB12Not started

Week 2 · Sensor + audio

  1. 06Sensor model: CNN vs TransformerDays 8–10 · Mon 2 – Wed 4 Nov · NB20 · NB21Not started
  2. 07Audio model with your own MFCCDays 11–12 · Thu 5 – Fri 6 Nov · NB30 · NB31 · NB32Not started
  3. 08One config drives everythingDays 13–14 · Sat 7 – Sun 8 Nov · NB90 v0Not started

Week 3 · Detection + gate

  1. 09Object detection (centroid)Days 15–17 · Mon 9 – Wed 11 Nov · NB40 · NB41 · NB42Not started
  2. 10Profiler and device simulatorDays 17–18 · Wed 11 – Thu 12 Nov · NB50 · NB51Not started
  3. 11Nemotron advisor and RAGDay 19 · Fri 13 Nov · NB60 · NB61Not started
  4. 12NVIDIA vision helpersDay 20 · Sat 14 Nov · NB62Not started
  5. 13Adoption gateDay 21 · Sun 15 Nov · NB90 finalNot started

Week 4 · Notebooks → web

  1. 14Notebooks to an ml_core packageDay 22 · Mon 16 Nov · Python and web codeNot started
  2. 15FastAPI and the training workerDays 23–24 · Tue 17 – Wed 18 Nov · Python and web codeNot started
  3. 16Next.js slice and demoDays 25–30 · Thu 19 – Tue 24 Nov · Python and web codeNot started

Before Day 1

  • Jetson Orin NX 16GB dev kit (Super), JetPack 6.2.1, with its original power supply and the fan working
  • NVMe SSD with 100 GB+ free (datasets, GGUF models, exports)
  • Laptop with an SSH client and VS Code or a browser
  • Phone with a voice recorder and the free phyphox app (sensor data)
  • Optional: USB webcam (drowsiness) and USB microphone (bird sounds)

Where everything lives

On the Jetson

  • All code, data, models and notebooks under ~/sutraai
  • JupyterLab runs here; you open it from the laptop browser
  • Python venv: ~/sutraai/.venv

On your laptop

  • SSH terminal + browser (or VS Code Remote-SSH)
  • Label Studio for drawing boxes (Step 09)
  • These pages and the docs: sutracore.com/learn/nvidia-jetson-ai
File~/sutraai (create in Step 02)
~/sutraai/
├── notebooks/        NB00 … NB90 — one notebook per step of the flow
├── sutra_core/      .py helpers the notebooks import (becomes ml_core in Step 14)
├── configs/          image.yaml  sensor.yaml  audio.yaml  detect.yaml
├── data/raw/         your photos, CSVs, WAVs, labels as recorded
├── data/processed/   train / val / test splits
├── models/           best.pt checkpoints, GGUF LLM files
├── exports/          .onnx  .tflite  .engine  model_data.cc  manifest_*.json  golden/
├── reports/          metrics.json, confusion matrices, profile.md
└── BUILDLOG.md       one entry per day: learned, built, numbers, blocked, tomorrow

Documents