Established NetKit Labs to develop embedded C/C++ software for MCU and embedded Linux platforms.

Website · Repositories · Project guide · Explore by topic

Our work spans bare-metal and Zephyr RTOS firmware on Arm Cortex-M/A, embedded Linux, audio DSP, TinyML inference runtimes, and FPGA-based neural network acceleration. The projects connect embedded software infrastructure with keyword spotting, speech commands, and efficient inference on microcontrollers, MPU hosts, and programmable logic.

• Embedded Software Development (C/C++)
• Arm Cortex-M/A Firmware
• Bare-Metal & Zephyr RTOS Systems
• Embedded Linux
• Audio DSP & Keyword Spotting
• Edge AI / TinyML
• Neural Network Inference Runtimes
• FPGA AI Accelerators
• Hardware/Software Co-Design

Built a C++20 keyword-spotting application for the Arduino Nano 33 BLE Sense (nRF52840 / Arm Cortex-M4F), using Zephyr RTOS, CMSIS-DSP features, and int8 TensorFlow Lite Micro inference. Continuous 16 kHz microphone audio drives voice-triggered LED control.

Repository: NetKit-Labs/kwskit

Highlights include:

• Separate thread contexts for PDM capture, audio processing, keyword control, and telemetry
• Fixed application buffers, thread stacks, and a 64 KiB inference arena
• Nonblocking score delivery, bounded audio batches, and stale-result rejection
• Unit and regression tests, firmware CI, and recorded hardware validation
• Spoken-keyword operation and physical LED toggling verified on the Nano

NetKit is a C++26 inference runtime with C++ and C23 APIs for MCU and embedded Linux targets. Float32 and INT8 paths support resource-constrained deployment, with optimized backends including CMSIS-NN, ESP-NN, NMSIS-NN, and XNNPACK.

Repository: NetKit-Labs/netkit

Highlights include:

• C++ and C APIs over a shared runtime
• Float32 and INT8 quantized inference
• Explicit memory planning for embedded deployment
• Optimized kernels for Arm, Espressif, RISC-V, and host CPUs
• Peer benchmarks against established inference runtimes

MemKit is an embedded C++26 container library with C++ and C23 APIs over shared implementations, designed for bare-metal MCUs, RTOS firmware, and embedded Linux MPUs.

Repository: NetKit-Labs/memkit

Features include:

• Ring buffers, maps, pools, and other embedded containers
• C++ lock-free queues and double buffering for task/ISR handoff
• Static and arena storage for MCU firmware
• Heap and virtual-memory arena backing for MPU hosts
• C and C++ APIs over shared container cores

NPUKit is a SystemVerilog INT8 neural network accelerator for the PYNQ-Z2 (AMD-Xilinx Zynq-7020). An 8×8 output-stationary systolic array performs matrix multiplication, with AXI control, DMA tile transfers, and transformer operations split between FPGA fabric and Python host software. Speech-command experiments explore embedded workloads on top of the accelerator.

Repository: NetKit-Labs/npukit

Highlights include:

• 8×8 INT8 systolic GEMM array
• AXI4-Lite control and AXI DMA tile I/O
• On-fabric residual, GELU, RMSNorm, and Softmax operations
• Python host software for board control and inference
• Hardware/software co-design with speech-command workloads

Languages

• C
• C++
• SystemVerilog
• Python

Platforms

• Arm Cortex-M
• Arm Cortex-A
• Zephyr RTOS
• Embedded Linux
• AMD-Xilinx Zynq SoCs
• FPGA

Domains

• Embedded Systems
• Audio DSP
• Keyword Spotting & Speech Commands
• Edge AI / TinyML
• Neural Network Acceleration

James Lavrenz

Embedded Software Engineer | C/C++ | Embedded Linux | RTOS | Arm Cortex-M/A | Edge AI

Website · GitHub · LinkedIn

For project questions and bug reports, use the issue tracker in the relevant repository: KWSKit, NetKit, MemKit, or NPUKit.