About

Background and approach

I am Kelvin Chandra Djayapranata, an undergraduate student at Northwestern Polytechnical University (NPU/NWPU) in Xi'an, China. I work on the data side of AI for Biotechnology (AI4Bio) — bioinformatics and computational biology applied before the modelling step: protein sequence and structure analysis, residue-level feature extraction, and validated, reproducible datasets built from AlphaFold predictions. This is a personal portfolio; it is not an official NWPU website and is not endorsed by the university.

Academic background

I am currently an undergraduate student at Northwestern Polytechnical University (NPU/NWPU), Xi'an, Shaanxi, China. My stated areas of interest are AI for Biotechnology (AI4Bio) and computational biology.

This is the full extent of what I publish about my education here. Program, year of study, honours, and awards are not listed because they are not part of my public record.

What I focus on

Most of my work sits before the modelling step. In computational biology the quality of a result depends on the quality of the data behind it — where it came from, what each column means, what units it uses, and what was deliberately left out. So my focus is on:

• Protein sequence and structure analysis using predicted models
• Residue-level feature extraction with fully documented definitions and units
• Building validated, reproducible datasets from public structure databases
• Understanding the confidence measures that come with predicted structures

How I approach the work

My public repositories document a consistent set of practices, and this is the approach I try to hold to:

• Recompute rather than trust. Integrity tests recalculate critical values from the raw prediction files instead of accepting derived outputs.
• Document definitions and units. Every feature column is defined in a machine-readable schema, so a number means the same thing to everyone reading it.
• Record exclusions explicitly. When data is unavailable, the column is excluded and the reason is written down — not imputed or silently redefined.
• Inspect before changing. Read-only review comes first; findings get reported rather than fixed automatically.
• Report the limits alongside the result. Every project states what it does not do, and null or negative outcomes stay on the record.

Where I am right now

I am early in this work, and I would rather say so plainly. My projects are at the pipeline and documentation stage — not at the stage of finished results. What exists today is validated data infrastructure, test suites, and technical writing. What does not exist yet is a trained model, a biological finding, or a publication.

If you are working in computational biology or machine learning and see something worth correcting or building on, I would genuinely like to hear it.