Process
How a project actually runs
Five phases, each ending in something reviewable. The names are ordinary; the discipline is in refusing to skip one because the schedule is tight.
- 01
Scope
Agree what the product must do and what "finished" means.
- Requirements written down — including the ones that are really constraints
- Volumes, target cost and certification path identified
- Risks and unknowns listed openly, with a plan for each
- A proposal with a scope, a price and an honest schedule
- 02
Architect
Decide the shape of the system while changes are still cheap.
- Block diagram, power architecture and interface map
- Silicon and module selection, lifecycle and availability checked
- Hardware / firmware / FPGA / model partitioning
- Early proof of the one or two things that could sink the project
- 03
Design
Build the thing — board, code and logic developed in parallel.
- Schematic capture, reviewed before layout starts
- PCB layout with signal integrity, EMC and thermals considered
- Firmware and RTL developed against the real requirements, in version control
- Design reviews at the points where they still change the outcome
- 04
Prove
Measure instead of assume.
- Prototype build, bring-up and characterisation
- Validation against the requirement list, with results recorded
- EMC pre-compliance and the design changes that come out of it
- A short report on what passed, what did not, and what we changed
- 05
Produce
Hand over something that can be built without us.
- Complete manufacturing data pack and assembly documentation
- Production test specification and, where needed, test fixtures
- Assembler selection, first-article review and yield follow-up
- Sources, documentation and design files handed to you
Principles
Four things we do not trade away
Written decisions
Every significant choice — component, architecture, trade-off — ends up in writing with the reason behind it. Two years later, when someone asks why, the answer exists.
Measured, not assumed
Datasheet numbers are a starting point. Margins, temperatures, timings, currents and inference latency get measured on real hardware before anyone calls the design done.
Bad news early
A slip we tell you about in week three is a planning problem. The same slip revealed in week twenty is a crisis. We would rather have the uncomfortable conversation first.
No lock-in
Standard tools, standard formats, complete handover. You should be free to continue with your own team or another supplier at any point.
Get in touch
Bring us in at the phase you are in.
Projects do not always start at the beginning. A review of an existing design, a rescue of a stalled one, or a second opinion before you commit to production are all normal starting points.