Embedded software for industrial, automotive and IoT products

Meplis has been developing firmware for more than eight years. We usually join a project at the first prototype board and stay until the devices are in the field and receiving updates.

Scope
Architecture, bring-up, production firmware, field updates
Architectures
ARM Cortex, RISC-V, C2000, TriCore
RTOS
Zephyr, FreeRTOS, bare-metal
Languages
C, C++, Rust
Edge AI
On-device ML, model quantization, sensor fusion
Connectivity
Wi-Fi, Ethernet, LoRa, BLE, cellular, CAN
Verification
HIL, SIL, MISRA-C, CI
24 V DC 31 % START J1 · ETH T1 PHY U4 Wi-Fi · BLE U3 CORTEX-M7 MAIN · NPU Y1 U1 RISC-V RT CORE Y2 U2 CAN J3 · CAN SE U8 QSPI J4 3V3 L1 D1 D2 D3 J2 · SWD meplis.dev REV 2026.11 EDGE CONTROLLER · MX-3 IP65 · CE REV 2026.11 PWRLINKRUN AUTOMOTIVE · CAN FD 0x18F 3A 00 12 7F C4 010x0C1 00 FA 20 100x2A4 11 03 7E 00 00 9B0x18F 3B 00 12 7F C4 020x7DF 02 01 0C0x0C1 00 FB 20 100x3E8 55 AA 010x2A4 11 03 7F 00 00 9C0x18F 3C 00 12 80 C4 030x101 08 000x18F 3A 00 12 7F C4 010x0C1 00 FA 20 100x2A4 11 03 7E 00 00 9B0x18F 3B 00 12 7F C4 020x7DF 02 01 0C0x0C1 00 FB 20 100x3E8 55 AA 010x2A4 11 03 7F 00 00 9C0x18F 3C 00 12 80 C4 030x101 08 00 INDUSTRIAL · MODBUS TCP L1231.4 VL2229.8 VL3230.6 VI12.7 AT41.2 °C ENERGY · POWER QUALITY THD 2.8 %PF 0.9750.01 Hz · 3-phase AGRICULTURE · LoRaWAN PUMP 1ON · 42 m³/hPUMP 2IDLESOIL31 %TANK68 %LINK-97 dBm SF9 WEARABLE · ANOMALY ANOMALY p = 0.97on-device · int8 · 3 ms

Edge AI

More of our projects now run a trained model on the device itself. The device reacts within milliseconds, keeps working without a connection, and raw sensor data does not have to leave it.

  1. On-device inference

    Models that run on the MCU or embedded processor itself, inside a fixed RAM, flash and power budget. No round trip to the cloud.

  2. Model optimization

    Quantization, pruning and operator-level tuning, so a model trained on a GPU fits the target and meets its real-time deadline.

  3. Signal pipelines

    Filtering, feature extraction and sensor fusion written as production firmware, not notebook code.

  4. Model lifecycle

    Data capture from devices, validation on hardware in HIL, and versioned model updates shipped over OTA with the firmware.

SENSOR STREAM FEATURES ON-DEVICE MODEL · INT8 normal0.94anomaly0.05fault0.01 RAMFLASHINFERENCE TIME

Typical uses: anomaly detection on motors and pumps, predictive maintenance, power quality analysis, keyword spotting and visual inspection.

Services

Firmware & Drivers

Drivers for SPI, I²C, UART, ADC and DMA, bare-metal or on top of the vendor HAL.

RTOS Architecture

Zephyr and FreeRTOS: task design, priorities, IPC and memory budgets that hold under load.

Connectivity & IoT

Wi-Fi, Ethernet, LoRa, BLE, cellular and CAN. MQTT and TLS on top, with reconnects and buffering for unreliable links.

OTA & Security

Signed images, A/B slots and rollback if the new firmware does not come up.

Test & Verification

HIL and SIL setups, host-based unit tests with mocked hardware, MISRA-C checks in CI.

Debug & Bring-up

SWD debugger, logic analyzer and scope, for bugs that are half hardware.

Infrastructure

We run our own MQTT brokers, OTA update server and LoRaWAN gateway for development and field tests. The console below shows a sample device log.

Probing services…

Get in touch

Tell us about the hardware, the requirements and the timeline.