Aleksandra Sharonova

I’m a Data Scientist and ML Engineer with 7+ years in data and 3+ years building ML in production. I work mostly with uplift and geospatial models, NLP and transformer fine-tuning, A/B testing, and retention—always with a focus on models that solve real business problems and actually get used.

At Uzum, I built a geospatial uplift model (CatBoost, SHAP) with a Kepler.gl map that opened 40 new pickup points in 6 months at about 25% MAPE, replacing manual expert picks. I also trained a fastText classifier for search queries and used it in an assortment dashboard to quantify GMV lost on in-demand products not yet in the catalog. For churn, I modeled churn drivers with CatBoost and SHAP and found distance from the customer to the pickup point among the top three reasons.

At Wakie, I built a matching model (logistic regression) that raised retention by 10% (up to 30% on secondary metrics). I developed a two-stage content moderation pipeline—starting with a TF-IDF and gradient boosting baseline, then moving to a fine-tuned RoBERTa—to handle tricky cases and cut manual review. I also forecasted promotion impact on user payments for A/B tests using ARIMA and then Prophet, achieving about 5–10% MAPE.

Earlier in my career, I supported the full analytics and engineering loop: moving Python ETL pipelines and databases from Heroku to AWS as a Data Engineer (contract), and driving experimentation and measurement as a Product Analyst and Team Lead. From 2018–2021, I built attribution models (Shapley, Markov chains) and raised conversion by 34% in a key segment, reinforcing my preference for outcomes that can be measured end-to-end.