2weeks
Discovery
An architecture, a build plan, a risk list and a manufacturing sanity check.

The offer
No open-ended retainers, no discovery phase that discovers itself. Every stage has a fixed shape, a fixed length and a deliverable you can hold, run, or hand to a manufacturer.
2 · 4 · 8 weeksDiscovery · Software · Hardware prototype
2weeks
Discovery
An architecture, a build plan, a risk list and a manufacturing sanity check.
4weeks
Working software
Inference on real hardware, the API, the data path and a client you can press.
8weeks
Hardware prototype
A prototype in its enclosure, the BOM and the documentation a factory quotes from.
STAGE 01
We take the idea apart with you: what has to happen locally, what can wait, which sensors and silicon, which model and how big, what the enclosure implies, what it costs to manufacture at the volumes you have in mind. You end the fortnight with an architecture, a build plan, a risk list and a manufacturing sanity check.
2 weeks
STAGE 02
Software running on real hardware, a development board or an existing unit, with inference on the target, the data path, the API and a client you can actually press. Early, unfinished and truthful: the point is to meet the problems now, while changing your mind is still free.
4 weeks
STAGE 03
A working prototype in its own enclosure, with the bill of materials and the complete documentation set a contract manufacturer needs to quote mass production. Printed parts designed to be moulded, an assembly sequence, a test plan, and an honest list of what is still open.
8 weeks
Stages run in sequence or stand alone. Plenty of clients arrive with hardware and need the intelligence, or with a model and need something to put it in. Pricing is quoted per engagement once we know the shape of the problem; discovery is the cheapest way to get there.
A prototype that cannot be manufactured is a sculpture. Beautiful, expensive, and no use to a sales team.
Fit
Good fit
A deadline in milliseconds, data that must not leave a building, a site with no reliable uplink, or a per-event cloud bill that gets worse as you grow. Constraints are what make local AI the cheaper answer rather than the clever one.
Good fit
Manufacturers modernising a proven line, teams whose cloud prototype just met its first real installation, founders who need one group to own hardware, model and software instead of three vendors pointing at each other.
Poor fit
If it runs happily in a browser and nobody cares about latency, you do not need us, and we will tell you so in week one rather than week nine.
Poor fit
Hardware added to make a software product look serious tends to produce an expensive object that does less than the phone already in the room. We would rather build you the thing that has to exist.
Week zero
Whether it can be built. What it costs to make. What has to be true for it to work. That is a cheap way to find out, and an expensive thing not to know.