High-frequency acquisition
Acquiring, processing and storing at frequencies up to 1 kHz, where the difficulty is timing rather than throughput.
A thousand samples a second is not a lot of data. Modern hardware moves far more without complaint, which is why throughput is the wrong thing to worry about.
What is hard is that every one of those samples has to be where it says it is in time, that none of them may quietly go missing, and that months later you have to be able to show that a given sample is what it claims to be.
Timing integrity
A sample with the wrong timestamp is not slightly wrong, it is wrong in a way that survives every downstream stage looking correct. Jitter has to be bounded and measured, not assumed, and the measurement has to be part of the record rather than something you check when you already suspect a problem.
The same applies to gaps. A pipeline that silently drops samples under load produces data that looks continuous and is not, and nothing downstream can detect that unless the acquisition layer says so.
Provenance
In a regulated context, the value of a measurement is bounded by what you can demonstrate about it. So the chain from the physical signal to the stored series carries its own evidence: what the sensor was, what the conversion was, when it was calibrated, which software version handled it.
That is a storage and schema decision taken at the start. Retrofitting provenance onto a series that was collected without it does not produce provenance, it produces a claim.
What I will not publish
The physical quantities, the instrument, the deployment, the sector detail beyond what is on this site, or any of the software. Described at a high level under NDA.
Work carried out under NDA. This page describes the engineering, not the employer. No employer names, internal architectures, client details or proprietary systems appear anywhere on this site, and none will be added.