Andi T.

Most people who ask "where is my ancestry actually from?" get a colored pie chart and no error bars. I give you a tested model: qpAdm admixture proportions with standard errors and p-values, every competing model that was rejected, and the code that reproduces all of it. I run formal admixture modelling on human genetic data — qpAdm, qpWave and the f-statistic framework (ADMIXTOOLS 2 in R) — and turn the output into something you can actually use: a readable report, a publication figure set, or a running pipeline. ── WHO HIRES ME ── 1) GENETIC GENEALOGISTS AND DNA HOBBYISTS You have raw data from 23andMe, AncestryDNA, MyHeritage, FTDNA or a WGS test and you want more than a consumer breakdown. I merge your kit with the Allen Ancient DNA Resource, run distal and proximal qpAdm models against real ancient populations, and send you a plain-English report: which models fit, which were rejected and why, your ancestry proportions with confidence intervals, and PCA / Global25 plots showing exactly where you fall among ancient and modern populations. 2) RESEARCHERS AND GRADUATE STUDENTS You need an analysis that survives peer review and a methods section you can defend. I build the whole chain — genotype QC and merging (PLINK, EIGENSTRAT / PACKEDANCESTRYMAP), f2 pre-computation, rotating-outgroup qpAdm screens across hundreds of candidate models, qpWave rank tests, nested-model comparison — and deliver ranked results, publication-quality figures, and a documented R/Python repository your co-authors can rerun line for line. 3) ANCESTRY AND BIOTECH COMPANIES This is where I differ from most analysts: I am a software engineer first. Ten years building production systems in C#/.NET, Azure and SQL before I moved into population genomics. If you need raw upload → QC → reference merge → model → served result as an actual product rather than a folder of scripts, I can design it, build it, document it and hand it over. I run exactly this stack for my own consumer ancestry platform, Ancestrify. ── TOOLING ── ADMIXTOOLS 2 and classic ADMIXTOOLS (qpAdm, qpWave, qpDstat, qp3Pop, qpF4ratio, qpGraph) · smartpca / EIGENSOFT · ADMIXTURE · PLINK 1.9 and 2.0 · Global25 with nMonte and Vahaduo · DATES for admixture dating · hapROH · R (tidyverse, ggplot2, data.table) · Python · SQL · Azure ── HOW I WORK ── • I tell you what your data can and cannot support. If coverage is too low for a distal model, you hear that before you pay for one. • Every result ships with its reasoning: p-values, standard errors, the outgroup set used, and the models that failed. • Everything is reproducible. You get the scripts and parameter files, not just a screenshot of the output. • 16 contracts and 635 hours delivered on Upwork, with clear and fast communication throughout. Send me what you have — kit type and coverage, the populations you care about, or the paper you are working toward — and I will tell you straight away whether qpAdm is the right tool for the question, and if it is not, what is.