To optimize their beamline, scientists want to catch as much photons as possible, using ideally area detector at high frequency (khz) even if they are interested in spectra (1D curves) or just a single value. My job is to reduce images into something meaningful for them... at the speed of the acquisition.
This implies three main steps:
* ability to understand the scientific problem and communicate with beamline scientists
* description of the algorithm needed with often a parallelization step (applied math)
* implementation of those algorithm into program/libraries running with real-time constrains inside the acquisition chain.