I work at a non-profit Center of Excellence owned by the national development fund, on a team of ten. The mandate is unusual for embedded work: digital sovereignty, green electronics through IoT, and digital transformation for Albanian public institutions. The hardware has to survive procurement cycles, not product cycles — nobody refreshes a municipal sensor fleet every eighteen months.

Self-sustained devices

The deployment I care about is the one nobody visits again. If servicing a node means a lift, a permit, and a technician, the battery isn’t a component — it’s a recurring liability, and across a fleet the maintenance schedule is the cost structure.

So the design inverts: start from the energy the environment actually offers and work backwards. Indoor PV, thermoelectric where there’s a stable gradient, piezo on rotating machinery. Two-tier storage — supercap for the radio’s pulse loads, cell for the slow baseline, because servicing pulses from lithium turns a ten-year design into a three-year one. Duty cycles budgeted honestly, down to what a single LTE-M attach costs. Adaptive wakeups so a cloudy week degrades gracefully instead of browning out mid-transmit. And a state machine that survives an unplanned reset without losing its calibration baseline — intermittent computing starts at the schematic.

Done right, the device outlives its cell and dies of component age instead.

R&D: smart city digital twins

The layer above the sensors — turning a fleet of nodes into a model a city can decide with. A twin is only as honest as its worst data source, so this isn’t separable from the hardware work: provenance, timestamp integrity, and drift at the node decide whether the simulation is decision-grade or a very expensive screensaver.

In practice: a 3D base model at whatever detail the question justifies, live state bound to physical entities through standardised context models, and what-if simulation as the actual point. Open formats throughout, because the model outlives the vendor and a small country can’t afford to fund the same model twice — let alone lease access to a model of its own streets.

What I stand for

Microamps over megahertz. Power is a day-one constraint, not a post-demo optimisation. Repairability is a layout decision — test points, labelled headers, no glue where a screw would do. Sovereignty over convenience: if your device dies when someone’s servers do, you never bought a product.