Selected work
05 / 08Embedded Linux · Edge AI2026

CASE STUDY · 2026

Custom OS Architecture & Edge Inference Engine.

Build a purpose built ARM64 environment for edge inference.

ARM64 software emulation prototype

A custom Buildroot Linux image for ARM64/QEMU, paired with a multithreaded C++ frame processing path and a software emulated MobileNet SSD inference pipeline.

OVERVIEW

What it is
and who it helps.

This project builds a small custom operating environment around an ARM64 target. A cross compiled Buildroot toolchain produces the root filesystem and kernel layout, while an asynchronous C++ application ingests downsampled frames and sends them through a MobileNet SSD path running in QEMU software emulation.

ARM64Target architecture
QEMUSoftware emulation
MobileNet SSDEdge inference path
THE CHALLENGE

The work had to join a reliable target image with a useful inference path despite running under software emulation. Cross compilation also surfaced target multimedia dependencies and toolchain optimization settings that needed to be resolved before frames could move cleanly through the system.

HOW I APPROACHED IT

From the first
interaction to the edge cases.

01

Shape the target operating system

Configured a cross compiled Buildroot toolchain and built the ARM64 root filesystem and kernel layout from scratch, keeping the image small enough for iterative emulation.

02

Make the toolchain reproducible

Resolved cross compilation failures by configuring target multimedia dependencies and the toolchain optimization variables expected by the target image.

03

Stream frames without blocking the pipeline

Wrote an asynchronous, multithreaded C++ ingestion engine that reads downsampled frame buffers over custom network sockets.

04

Run inference at the edge

Integrated a MobileNet SSD neural network path in the emulated micro OS and exported the resulting detection boundaries for downstream use.

OUTCOME

A full edge path inside a custom OS.

The prototype brings a custom Linux image, asynchronous frame transport, and object detection inference into one ARM64/QEMU workflow. It makes the target system and its software path testable before moving to physical edge hardware.

BUILT WITH

BuildrootLinux KernelARM64 / QEMUC++MobileNet SSDCross compilation

UP NEXT · 06

Fault Tolerant IoT Telemetry Stack.

Read case study