Hardware & Embedded
Edge AI
Machine learning that runs on the MCU. We work with TensorFlow Lite Micro and CMSIS-NN to quantise models, profile their footprint, and fit them into the RAM and flash you actually have. Useful when latency, privacy, or offline operation rule out a cloud round-trip — anomaly detection on vibration data, keyword spotting, simple computer vision on grayscale frames.
What we offer
Capabilities
Model quantisation: INT8, INT4, mixed-precision
TensorFlow Lite Micro on Cortex-M and ESP32
CMSIS-NN kernels for ARM targets
Sensor preprocessing pipelines: windowing, FFT, MFCC
Memory and latency profiling on the actual silicon
Model retraining loops driven by field data
Tech stack
What we reach for
TensorFlow Lite MicroCMSIS-NNEdge ImpulseONNX RuntimePyTorchTensorFlow
Our process
How we deliver
01
Frame the problem
What signal, what decision, what latency budget.
02
Model & quantise
Train in PyTorch or TF, quantise to fit the MCU.
03
Profile
Real silicon, real RAM and flash, real latency.
04
Iterate
Retrain on field data; ship updates via OTA.
Talk to us
Interested in this service?
Tell us what you're building. We'll let you know whether it's a fit, and where it isn't.