When X-rays scatter from a crystal or a thin film, a detector records intensities—not a ready-made image of the structure or a list of layer thicknesses. Those have to be worked out from the measurements, and sometimes more than one answer appears plausible.

I build the software for that work: predicting diffraction patterns, fitting X-ray reflectivity curves, reconstructing images, and turning raw beamline data into quantities we can compare. Most of my research code is Python (NumPy, JAX, PyTorch); I also explore C++20/CUDA in QoraPhaseCU.

• TraTT: diffraction while sound enters a crystal. A sound wave can produce extra peaks beside the main Bragg peak. I simulate how those peaks form before the wave has crossed the crystal, using time-dependent Takagi–Taupin equations. Calculated findings · KPS 2026 Fall poster.
• ReflectoLearn: film structure from X-ray reflection. An X-ray reflectivity curve contains information about film thickness, roughness, and scattering strength. I use a neural-network starting guess and a physics-based fit, then study when that estimate goes wrong or cannot be distinguished from another structure. Findings and limits.
• QoraPhase: images from diffraction intensities. Diffraction measurements lose the wave's phase. This JAX package iteratively estimates the missing phase and the object that produced the pattern. Synthetic demonstration · Measured-data limitations.
• QoraFlow: raw detector frames to time traces. At PAL-XFEL, a laser excites the sample and X-rays probe it after different delays. The pipeline groups detector shots by laser state and delay, corrects the images, and tracks changes in a selected diffraction peak. Processing steps · Simulated example.

These are recorded calculations and benchmarks with specific limits—not claims that every program has been validated on an experimental sample.

Read the evidence and reproduction guide — sources, commands, data-access limits, and the next validation needed for each claim.

_{Not measured data: a simulated diffraction intensity reconstructed with QoraPhase. Reproduce the animation.}

• Numerical performance: QoraPhaseCU explores C++20/CUDA phase retrieval; xray.ooguy provides beamline calculators.
• Scientific figures: qoraplot supplies Matplotlib publication layouts; cvdkit pairs color-vision-aware palettes with markers and line styles.
• Developer workflows: codigest packages selected code/diffs/terminal output as AI context; herdr-courier is designed to keep agent prompts from interrupting typed human input; hanfix repairs terminal commands typed under the Korean keyboard layout.

Technical notes · Repository index (last audited 2026-10-23)

_{Pepper (후추), my black-and-tan Shiba Inu, takes care of the contribution garden.}

_{Formerly "SJB7777". Renamed to "NaNIsaac" on October 14, 2026, at 00:06:41 KST.}