Machine-to-Machine Translation
Two machines that would not talk, each with its own coordinate system. I built the translation between them, and the lab put it straight into their software.
A medical research lab ran a two-machine process: a specialist marked the regions that mattered on one machine, and a second machine then acted on those exact regions. Carrying the marked regions from the first machine to the second was approximate and took real mental effort, and the targeting was only ever as good as that translation. The method I built made the handoff automatic and precise. The regions land exactly where the specialist marked them, with no one burning brainpower to re-derive them by hand. It ran reliably, and held up so well they built it straight into their commercial software and brought me back again years later.
In his wordsOne of the first projects he did for us was getting two of our lab instruments to talk to each other, which nobody had a good way to handle at the time. He built an automated process that solved it which was also later packaged into our commercial software products.
Two machines, two coordinate systems, and a sample that could flip
On the first machine, a specialist read a sample and marked the regions that mattered as coordinate shapes. A second machine then had to act on those same regions. The two machines used different coordinate systems, different scales and increments, so a region marked on the first did not map directly onto the second.
Worse, the regions were not simply rotated and scaled every time. Someone handling the sample in the lab might flip it over between machines, and when that happened the regions had to be mirrored into a new coordinate system before anything else would line up. Getting the second machine onto the right regions was approximate, and the work was only as precise as that translation.
Three reference points, and the data settles the flip.
A method that took the regions marked on the first machine and re-derived them in the second machine's coordinate system, automatically. I programmed it from the ground up in Excel, which was the one channel both machines could read from and write to, so it became the bridge between them.
The method works from three reference points, call them 0, 1, and 2. The first version I wrote was heavy: a stack of trigonometry and conditional checks that rotated, scaled, and tested the geometry to decide whether the sample had flipped. It worked, but it was more complicated than the problem needed. So I simplified it. Instead of proving the flip with math, I let the data settle it: anchor on points 1 and 2, then work out where the third point should land under both assumptions, flipped and not flipped. Compare each prediction against where that point actually sat in the second machine's data, and take whichever assumption has the smaller error. Once that settles whether to mirror, the rest is a clean scale and rotation, and the whole set of regions re-derives from the raw data into the second machine's plane.
The translator across the gap
The hard part was never the code. It was reducing a messy physical problem, two machines, two coordinate systems, a sample that can flip over, down to an order of operations that always holds: resolve the flip first, then scale and rotate.
Then the same instinct got tested a second way. The lab brought on a programmer to build the method into their commercial software. That programmer could build almost anything, in code and down at the hardware level, but the spatial math did not land for him. So that engagement was not handing off code. It was making the geometry legible, step by step, until it shipped inside the product. The first time I translated between two machines. The second time, between two people. They brought me back a third time, years later, for another translation entirely.
I am not loyal to any one tool or trade, and that is the point. The real capability here isn't the lab or the science, it's making separate systems talk to each other, and making the logic legible to whoever has to carry it next. This is the one piece of work that was fully my own, from start to finish. I built it once, and it kept doing its job long after I was gone.
Have something that looks like this pattern?
Free. 15 minutes. I'll know in the first 5 if I can help.