Is the answer you need not published anywhere yet?
Feasibility studies, prototypes, measurement and modelling, from embedded hardware to trained models: work whose outcome is knowledge rather than a product. The deliverable is a defensible answer, including when the answer is no.
The problem
A decision is waiting on a question nobody can answer from experience: whether a sensor can resolve the signal at all, whether a controller can hold a process within its limits, whether a model can tell two things apart reliably, whether an existing board does what its documentation claims. Without research the decision gets made on optimism, and the expensive version of the experiment happens in production.
How we approach it
We write the question down precisely, including what result would settle it either way, before any building starts. Then we design the cheapest experiment that can produce that result: a prototype on embedded hardware, a measurement campaign, a trained model, a numerical simulation. We run it with enough rigour that the numbers can be trusted, with calibration, controls and uncertainty stated. Where the work meets electronics we design the circuit, the antenna or the sensor chain ourselves, and we can take an existing board apart to find out what it really does. The report says what was found, how confident we are, and what it would take to be more confident.
Worked example
A growing system that reads the plant, not a timer
An early piece of our embedded work was a growing system that kept a plant in its optimal conditions instead of on a schedule. Sensors read humidity, temperature and the level of ultraviolet light, and a controller compared them against a profile for the plant species and drove the lamps, ventilation and watering to close the gap. The interesting part was not the hardware but the profile: finding which readings actually predicted growth, and how far each could drift before it mattered. Later work followed the same method in other directions: a vision model trained to tell two firearm types apart from photographs, antenna design and reverse engineering during industrial placements, and the analysis of circuit boards whose documentation had not survived.
What you get
- A precise statement of the question and what result would settle it
- The experiment, prototype, or model, built to a standard the numbers can bear
- A report with findings, uncertainty, and the next step, whichever way it points
- The data, code, schematics and instruments, so the work can be repeated or extended