Projects we can describe without naming the customer.
On a measurement product we worked on, a processing step after each run took about a minute. The hardware was fine. Operators were waiting on software.
We profiled that path, found where the time went, and changed it. Output and UI stayed the same. Runtime went from roughly 60 seconds to under 0.2 seconds, which made some follow-on features practical again.
On a sorting line there is only a short gap between the sensor reading and the actuator. If the decision is late, material is mis-sorted. If you add too much margin, the line slows down.
We built the decision path to stay under 10 ms end to end and checked it under load. The usual desktop patterns (blocking I/O, unexpected allocations, heavy middleware) were not a good fit.
Together with a hardware and installation partner we built a street-lighting system. Each lamp has a custom device with DALI dimming and schedules based on GPS sunrise and sunset. If the network is down, the lamp still follows its schedule from the device clock.
Because streets are long and thin, we used LoRa for longer links and WiFi mesh for shorter ones. Control moved from a scripted setup into a native driver. Issues with signal, topology, weather and power showed up in the field more than in the lab.
Lab Upgrade is a young company. The people behind it have worked in these areas for a long time.
A short review of the code, a talk with the people who use it, then a written note on what we would do next.
