i worked on this research during my summer as a silkroad fellow. the project studied ai readiness among 472 startups across five markets: kazakhstan, uzbekistan, kyrgyzstan, tajikistan, and azerbaijan.
my part was the engineering behind the data. the survey produced more than 700 records, and before anyone could compare them, they had to be consistent. i built a reproducible python and pandas pipeline that normalized, validated, and transformed those records into an analysis-ready dataset. in plain terms, every record ended up in the same shape and was checked before it reached the analysis.
the point of all that was simple: researchers could work from one consistent dataset, compare responses across startups and markets, and analyze ai readiness without cleaning the data by hand first. and because the pipeline is reproducible, running it again on the same records produces the same dataset.