Work queues in Zephyr RTOS are a lightweight way to defer work, move processing out of interrupt context, and keep applications responsive without creating extra threads. In this article, we explain how work queues work, explore system and custom queues, show how to schedule delayed tasks, and demonstrate a real-world Bluetooth advertising example where work queues ensure safe and efficient operation.
WS2812 LEDs, also known as NeoPixels, are popular for creating colorful lighting effects with just a single data line. The challenge lies in their strict timing requirements, which can be tricky to handle on microcontrollers. In this post, we’ll show how to control a WS2812 LED strip on an nRF52832 custom board using the nRF Connect SDK. By repurposing the SPI driver to generate the precise waveforms, we can drive the LEDs reliably without bit-banging.
Build configurations in the nRF Connect SDK let you fine-tune your application at compile time, controlling everything from logging verbosity to Bluetooth Low Energy parameters. In this post, we explore how to use prj.conf and menuconfig to manage Kconfig options, highlight common configurations such as logging, peripherals, and BLE settings, and walk through a practical example of enabling debug logging and tuning BLE buffers for higher performance.
Device Tree in nRF Connect SDK is a structured way of describing hardware, making your applications more portable and maintainable. Learn about DT hierarchy, nodes, properties, overlays, and practical examples like blinking LEDs, reading buttons, and adding custom sensors.
Getting started with the nRF52832 in the nRF Connect SDK requires understanding how Zephyr’s build system works. This post introduces the core components of building applications, including CMake, west, and key configuration files like prj.conf and CMakeLists.txt. We walk through how west orchestrates the build process, how to configure project features, and provide a step-by-step example of building and flashing a simple Hello World application. By mastering these fundamentals, you’ll be ready to develop scalable and maintainable embedded applications with Zephyr.
Environmental anomaly detection brings together the power of TinyML and microcontrollers to monitor conditions like gas levels, temperature, and humidity in real time. Instead of relying on fixed thresholds or labeled fault data, these systems learn what “normal” looks like and flag unusual patterns that could signal a problem — from gas leaks to rapid temperature spikes. By using lightweight models such as autoencoders, you can deploy intelligent, unsupervised anomaly detection directly on low-power devices, making it possible to create proactive, autonomous monitoring solutions for industrial, agricultural, and environmental applications.
With a simple accelerometer and the ESP32 WROVER, TinyML can turn everyday movements into powerful commands. By collecting and labeling motion data, training a lightweight neural network, and running it directly on the microcontroller, this project brings intuitive gesture-based control to smart devices without relying on the cloud. From wearable tech to robotics, the possibilities for touchless, natural interaction are endless.
Using TinyML on the ESP32 WROVER, a simple PIR sensor can become a smart motion detector that not only senses movement but classifies it in real time. By collecting diverse motion data, training a compact neural network, and deploying it efficiently on the ESP32, this system can distinguish between humans, pets, and background activity. With careful calibration, sensor placement, and optional Wi-Fi integration, it becomes a reliable, intelligent solution for home automation, security, and energy-saving applications.
The ESP32 WROVER kit, combined with common Arduino-compatible sensors, offers an accessible way to explore TinyML and bring real-time AI to embedded projects. From a simple voice-controlled RGB LED to advanced vision-based traffic detection, these project ideas progress in difficulty while introducing new skills such as audio preprocessing, sensor fusion, anomaly detection, and low-power optimization. By following this path, you’ll build a diverse portfolio of intelligent systems and gain hands-on experience in deploying machine learning models directly onto microcontrollers.
Explore how TinyML allows microcontrollers like ESP32 to recognize voice commands and control hardware like RGB LEDs, opening doors to low-power AI applications.
A voice-controlled toy car is a practical and engaging way to explore TinyML. By training a lightweight machine learning model to recognize specific commands and deploying it to a microcontroller, you can make a car respond to your voice without cloud services or external modules. This guide walks through the complete process from data collection to deployment, showing how to integrate AI, embedded hardware, and robotics into one project.
Voice control in embedded systems can be implemented in two main ways: running a custom TinyML model directly on your microcontroller or using a dedicated voice recognition module. Each approach offers distinct advantages in flexibility, performance, and development effort. This post explores how both methods work, compares their strengths and limitations, and helps you decide which is best for your next project.
TinyML is transforming the way AI interacts with everyday devices by enabling machine learning models to run directly on microcontrollers and other resource-constrained hardware. While much attention goes to deployment and inference, the training phase is where a model’s real capabilities are forged. Understanding the training process, from dataset preparation to optimization for embedded devices, is essential for building high-performance TinyML solutions.
TinyML is bringing artificial intelligence to the smallest of devices — microcontrollers with only kilobytes of RAM and ultra-low power budgets. At the heart of this movement are specialized software stacks that bridge the gap between cloud-trained models and the realities of embedded hardware. From TensorFlow Lite for Microcontrollers and Edge Impulse to CMSIS-NN and Nordic Semiconductor’s newly acquired Neuton platform, these tools provide the optimization, runtime efficiency, and hardware integration needed to run AI at the edge. This post explores the leading TinyML stacks, how they work, and how developers can choose and combine them for maximum impact.
TinyML brings machine learning out of the cloud and into the smallest of devices, enabling real-time, low-power intelligence at the edge. At the heart of this capability lies inference — the process of turning raw sensor data into actionable insights directly on a microcontroller with kilobytes of RAM and milliwatts of power. This article explores how inference works on resource-constrained hardware, the optimizations that make it possible, and the challenges developers face when balancing accuracy, performance, and efficiency.