AI for devices and operations

AI is useful when it answers a specific question with the data you have. Will this motor fail soon? Is the right item on the shelf? What did the operator ask for? We start from that question, then decide whether the answer runs on the device, in the cloud or in an app.

Book a discovery call

What we can take on

Machine monitoring that learns what normal looks like for each machine and flags changes.
Camera-based checks: counting, recognising items, detecting presence.
Voice and chat assistants for front desks and shop floors, in more than one language.
AI agents that prepare work for a person to approve, such as a report or a follow-up email.
AI on the device itself, from an FFT on a microcontroller to models on NVIDIA Jetson or Google Coral.

Machine monitoring and predictive maintenance

We record a baseline for each machine at its normal load and speed, then watch for changes against that baseline before any fixed threshold. Vibration analysis can start on the sensor itself, with only the results sent to the cloud. If the honest answer is that a simple rule on the device will do, we will say so.

Vision

Camera checks work when the lighting and the camera position are fixed first. We start there, then choose a model small enough to run where the answer is needed, on an edge computer next to the camera or in the cloud.

Voice, chat and AI agents

Voice and chat assistants answer the same questions every day so your staff do not have to, and hand over to a person when they should. AI agents go further and prepare work, such as a report, a summary or a reply, for a person to check and approve. We build and use these tools in-house, and run pilots for clients who want to try them on their own work.

For your engineers

Parts, protocols and tools we use for this work. The full list is on the technologies page.

FFT on microcontrollers
NVIDIA Jetson
Google Coral
YOLO-class detection
Camera and load-cell fusion
Vibration and temperature baselines
Multilingual voice AI
Claude and Gemini APIs
AI agents with human approval

Questions about this

Do we need a lot of data to start?

Not always. Often the first step is collecting the right data from one machine or one camera for a few weeks. That tells you whether the signal you need is there before anyone trains a model.

Should the AI run on the device or in the cloud?

It depends on how fast you need the answer, how much data there is and what the network can carry. Simple analysis such as an FFT runs well on the sensor node; heavier models run on an edge computer or in the cloud.

Have you deployed AI agents for clients?

We build and use AI agents in our own work every day, for reports, research and follow-ups, always with a person approving the output. Pilots for clients are available.

Tell us what you are building.