AI & Vision Modules
Pre-trained and fine-tunable models for industrial vision tasks.
We build vision systems that run where the camera is: weapon detection in public spaces, cell recognition under a microscope, PTZ perimeter tracking on a moving vehicle. The modelling work differs every time. The scaffolding around it barely changes.
These modules are that scaffolding, with a usable model already in it. Detection, anomaly spotting and an inference runtime that targets ARM rather than a datacentre GPU. Each one is meant to be fine-tuned on your data rather than used as-is — the value is that the pipeline, the pre-processing and the deployment path are already solved when your dataset arrives.
Object detection module
YOLO-class detection, retrainable on your own annotated set.
Anomaly detection engine
Learns a normal and flags deviation, for cases with few positive samples.
Edge inference runtime
Runs on ARM targets, with quantisation and batching handled.
Training pipeline
Dataset in, evaluated weights out, repeatable as data grows.
Annotation workflow
The unglamorous part, with the tooling and format already decided.
Camera abstraction
One interface across USB, MIPI, GigE and RTSP sources.
Frame the task
What counts as a detection, and what a false positive costs you.
Fine-tune
The module trains on your data rather than starting from nothing.
Evaluate
Measured against your acceptance threshold, not a public benchmark.
Deploy to the edge
Quantised and packaged for the target hardware.
Defense & Aerospace
Threat detection and tracking where the decision has to be local and immediate.
Medical & HealthTech
Image classification tasks with small, expensive, expert-labelled datasets.
Industrial IoT
Inline inspection that has to keep pace with the line.
