Zephyr Project Meetup – Dublin, Ireland
Welcome to the Zephyr Project in-person meetup in Dublin, Ireland, on October 29, 2026, from 5:00PM - 9:00PM.
This meetup is hosted by Analog Devices.
This gathering is for anyone interested in open source, embedded systems programming, and software development. Join developers, engineers, students, and open source enthusiasts for a morning of technical learning and networking centered around the Zephyr RTOS ecosystem.
As the Zephyr Project celebrates its 10th anniversary, this meetup offers an opportunity to connect with fellow community members, hear from industry experts, and explore how Zephyr is used in embedded and edge computing applications.
Technical Sessions
The evening will feature technical presentations covering open source development, system design, emerging technologies, and best practices for building connected devices with Zephyr. Learn from Zephyr contributors, ecosystem partners, and industry experts as they share practical applications and development experiences.
Whether you are an experienced embedded developer or just getting started, this meetup offers an opportunity to learn, exchange ideas, and connect with others working with Zephyr.
Seats are limited, so we encourage you to register early to secure your spot.
*Note: This meetup is part of the Zephyr Community Meetup Series, local gatherings hosted by community members with support from the Zephyr Project. *
Registration
Registration is required for this meetup so the hosts can plan the space and logistics accordingly.
Interested in sponsoring giveaway boards or food and beverages for an upcoming Zephyr meetup? Fill out the sponsorship interest form, and our organizers will contact you with event details and coordination information.
For questions related to meetup, please contact: meetups@zephyrproject.org Join the Zephyr Project Discord server (https://discord.com/invite/Ck7jw53nU2) to join the community discussions and post in the events-general channel if you have any event specific questions.
Contact organizers
-
5:00 PM - 5:25 PM GMT in-person
Welcome Snacks & Drinksin-person
-
5:25 PM - 5:30 PM GMT in-person
Welcome Note - Analog Devices Teamin-person
-
5:30 PM - 6:00 PM GMT in-person
Where Did My Message Go? Robust IPC on zbus for a Medical Wearablein-person
Many Zephyr-based embedded devices consist of independent subsystems running in parallel, including sensors, storage, logging, connectivity and power management, and they all need to talk to each other reliably. zbus is a solid foundation for that, but out of the box it's a publish/subscribe bus, not a complete IPC system. This talk presents the architecture of an IPC layer built on top of zbus for a medical wearable project. The layer includes a command/response model with matched replies and timeouts, broadcast events for system-wide state, and data pipes for streaming bulk data. Although it was designed for a medical device, the patterns apply to any multi-component Zephyr application that needs reliable IPC.
SPEAKERS -
6:05 PM - 6:35 PM GMT in-person
Is Zephyr on your radar?in-person
Zephyr supports radars that do their own processing and report targets, but not the growing range of FMCW mmWave radars that stream raw ADC samples for the MCU to process. Today every project using one writes its own driver and hard-codes chirp settings and antenna layout into its signal processing, so that processing only works with one chip.
At Beara, we are building a privacy-preserving assisted living device using sensor fusion around a mmWave radar and a ToF sensor, and we've proposed a common API to close the mmWave streaming gap in Zephyr. mmWave drivers stream raw frames through RTIO and describe how they were captured: chirp timing, frequencies, antenna positions and sample layout. Processing configures itself from the radar instead of from hard-coded constants; the chirp profile lives in devicetree, and the build fails if it doesn't match the register list that programs the chip.
This talk walks through the design of our open source BGT60TR13C driver, and walks through the pipeline we run on top of it; range-Doppler processing, target detection, 3D angle estimation, multi-target tracking and breathing detection. We'll finish with the question we'd like the community's input on before an upstream RFC: should raw radar frames get their own driver class, or build on the sensor API?
Attendees will take away:
- Why radar and ToF sensors suit assisted living, where cameras raise privacy concerns
- How raw mmWave radar data differs from typical sensor data, and why it needs its capture configuration alongside it
- Using RTIO to stream large frames, and devicetree to check a vendor configuration at build time
- The trade-offs between a new driver class and extending the sensor API
SPEAKERS -
6:40 PM - 7:10 PM GMT in-person
The horrors of Zephyr upstreaming in-person
Upstreaming is at the heart of the Zephyr Project, and the ideal process is well documented. But what happens when things don't go according to plan?
An overview of what can go wrong and what can you do about it when contributing patches to Zephyr upstream.
SPEAKERS -
7:10 PM - 7:40 PM GMT in-person
Break Timein-person
-
7:40 PM - 8:10 PM GMT in-person
MLForge: an AI agent in the Zephyr build loopin-person
Embedded ML models are usually trained for accuracy first and checked against the hardware last. That's when RAM, flash and latency turn up as surprises, and every surprise costs a retrain.
MLForge is an open-source proof of concept that reverses the order. You describe the target in one YAML file: board, tensor arena, RAM, flash, latency deadline and error target. An AI agent proposes a Keras network, and the pipeline trains it and quantises it to int8 TFLite. It then builds real TFLite Micro firmware with
west, runs it under QEMU or on the board, and feeds every measured result back into the next proposal, failures included.Zephyr makes this practical, because the same application builds for
qemu_cortex_m3and for real hardware. Cheap iterations run in emulation, and board time goes on the numbers that matter. The talk covers how the loop is put together, how it measures latency and error honestly (error bars, noise floors, on-device scoring), and what we measured on the FRDM-MCXN947. It also covers the Zephyr lessons we learnt the hard way: CMSIS-NN running 18 times faster than the reference kernels, silent console truncation, float formatting inprintk, and why the Neutron NPU path is built but still blocked.It opens with a short aside on where this started: detecting elephant trumpets on a solar-powered ESP32 recorder (ELOC). The talk ends with a live demo on the NXP board.
SPEAKERS -
8:15 PM - 8:35 PM GMT in-person
PyTorch to ExecuTorch: Deploying ML to Microcontrollers with Zephyrin-person
A few months ago I contributed the TensorFlow path for this in the form of the Edge Impulse Zephyr module SDK, and I've just had my first small PR merged into ExecuTorch itself.
This talk revisits the same deployment with the new ExecuTorch Zephyr module. ExecuTorch is PyTorch's on-device runtime, and it can now be pulled into a Zephyr build as a West module with a single Kconfig option.
The target is the Arduino UNO Q, a dual-architecture board with an STM32U585 Cortex-M33 running Zephyr alongside a Qualcomm QRB2210 Linux core, comparable to pairing a Raspberry Pi with the original UNO.
I'll walk through, as simply as possible, how to train a small PyTorch model, export it to a .pte, and run it on the MCU under Zephyr in tens of microseconds per inference, with the Linux side consuming the results over the internal UART without ever sitting in the real-time path.
The example is open source and reproducible with west init, so attendees leave with a working template for putting a PyTorch model on any Zephyr board.
SPEAKERS -
8:35 PM - 9:00 PM GMT in-person
Closing Note & Networkingin-person
SPEAKERS
No sessions match the selected labels.