Release status: Enrollment is open. The protected evaluator independently rebuilds the submission and the ESP32-S3 reference profile passes embedded self-test, replay parity, and malformed-input handling in QEMU.

This 18–22 hour course is for intermediate firmware developers who already work comfortably with C/C++, Python, and embedded build systems. Introduction to TinyML is recommended but is not an enrollment gate. Physical ESP32-S3 and BME280 hardware is optional enrichment only.

The course preserves the owner-supplied article, scripts, data, processed arrays, scaler, PyTorch checkpoint, ONNX model, metrics, plots, and 27-slide lecture deck as its foundation. Corrected artifacts sit beside the originals with explicit provenance.

What you will build

You will move one sensor history through a firmware-grade evidence chain:

  1. Treat Edge AI as a host/device firmware system.
  2. Audit the original BME280 stream and produce a dataset quality report.
  3. Create chronological, leakage-safe splits, scaling, and time windows.
  4. Train and evaluate the original compact LSTM against persistence.
  5. Export fixed-shape ONNX opset 18, verify it on the host, and convert it to a quantized ESP-DL artifact.
  6. Build real ESP-IDF/ESP-DL firmware, replay recorded vectors in ESP32-S3 QEMU, and collect UART evidence.

Four required labs lead to the Predictive Sensor Node capstone. Machine checks run first; final capstone approval requires instructor review. Successful learners receive an individual certificate whose number begins with EV-EDGE-AI-S3.

Lab environment

Completion uses three required environments: the browser sensor-window explorer with ONNX Runtime Web, a pinned GitHub Codespaces workspace, and Espressif’s ESP32-S3 QEMU running the real firmware image. The course never presents browser or host inference as MCU inference.

QEMU can establish firmware boot, deterministic recorded-vector handling, ESP-DL execution, error paths, and structured UART output. It cannot establish physical BME280 I²C behavior, real-time latency, energy consumption, radio behavior, or electrical correctness. Those remain optional hardware investigations and cannot affect completion or certification.

Price and access

The release price is $20 USD once through EmbeddedVille’s Skool checkout. A signed purchase event activates the matching verified EmbeddedVille account and creates the course enrollment idempotently. Premium access and administrator-reviewed manual verification remain available as fallbacks. No additional payment processor or course-specific account system is introduced.

Enroll in the course if you have already purchased. If you have not, purchase Time-Series AI on ESP32-S3 on Skool, then sign in with the matching verified email. EmbeddedVille never trusts learner-supplied receipts for automatic approval.