Data Scientist: market

Vacancy

an open job that cleared the quality gate and lists at least two technologies

Observed candidate

a candidate whose stack contains this technology; an aggregate — not a single record leaves the perimeter

Matchable candidate

not counted yet

Window

jobs open at the moment the snapshot was built

Junior DS — the most accessible entry into the ML/AI direction (typical path: Analyst 1-2 years → Junior DS). The Junior → Middle jump — after the first production model + first independent A/B test. Middle → Senior — multi-project ownership + business impact ownership (revenue / retention / conversion lift documented). Senior → Lead / Principal — DS strategy for a product area + hiring 3-7 DS + mentoring. Career flow: Data Analyst (1-2 years) → Junior DS (1-2 years) → Middle (2-3 years) → Senior → either Lead / Principal DS (deep), ML Engineer pivot (more technical track), Product Manager pivot, or Founder.

Top 5: python, sql, cloud, llm, aws. Python — monopoly (R has almost disappeared from new 2026 jobs, remaining in healthcare/biostats legacy). SQL mastery — mandatory: complex queries with window functions + CTEs + JOIN mastery + query optimisation (DS reads execution plans). Data manipulation: pandas (core), NumPy (numerical), Polars (modern, rising — 10-100× faster for big data), PyArrow. Classical ML — the core arsenal: scikit-learn (model selection / preprocessing / pipelines / metrics), XGBoost + LightGBM + CatBoost (gradient boosting — workhorse for tabular data), statsmodels (regression + time series + statistical tests). Statistics: scipy.stats, statsmodels, causal inference (DoWhy, EconML — Senior bonus). Visualisation: matplotlib + seaborn + plotly (in-notebook), Tableau / Looker / Power BI / Apache Superset / Metabase (for stakeholders). A/B testing: scipy.stats / statsmodels for analysis, internal platforms or Statsig / Eppo / Optimizely / GrowthBook. Understanding of power analysis, sample size calculation, multiple testing correction, sequential testing. Deep Learning (bonus, not mandatory): PyTorch / TensorFlow for CV / NLP DS tasks. Experiment tracking: MLflow or Weights & Biases. BI tools / dashboards: Tableau / Looker / Apache Superset / Metabase / Mode / Hex. Notebooks: Jupyter + VS Code (the 2026 market is migrating from raw Jupyter to hybrid). Cloud DS workbenches: AWS SageMaker Studio / GCP Vertex AI Workbench / Databricks / Snowflake Snowpark. Senior extras: causal inference (DoWhy / EconML / CausalImpact), uplift modelling, Bayesian modelling (PyMC / Stan), time series (Prophet / NeuralProphet / statsforecast).

Data Analyst — focus on reporting + dashboards + ad-hoc analysis. SQL + Excel + Tableau / Looker / Power BI. Less ML-heavy. Fits entry into the data direction. Data Scientist (this page) — focus on business problem framing + statistical modelling + A/B testing + insights communication. SQL deep + Python + sklearn + XGBoost + statsmodels + visualisation. Does ML, but doesn't own production infrastructure (deployment / serving / monitoring — that's the ML Engineer). ML Engineer — focus on production ML pipelines + deployment + scaling + LLM/RAG infrastructure. Stack DS + Docker / K8s + MLflow + Airflow + cloud-managed ML. See ML Engineer. Career flow: Analyst → DS Junior → DS Middle/Senior. DS ↔ ML Engineer pivot in 3-6 months (learn MLOps + Docker / K8s + cloud ML). DS ↔ Product Manager pivot in 2-4 months (DS already owns product thinking + A/B testing — add roadmap planning + stakeholder management). DS → Research Engineer / Scientist — requires PhD or 3+ years of deep specialisation.

Decision tree for picking the DS toolkit in 2026: 1) Tabular data, classification/regression, <1M rows → scikit-learn (LogisticRegression / RandomForest / GradientBoosting). Pipelines + preprocessing — sklearn wins for simple integration. 2) Tabular data, >100K rows, performance critical → XGBoost / LightGBM / CatBoost. The standard 2026 workhorse for tabular tasks: credit scoring, churn prediction, conversion, fraud detection, demand forecasting. LightGBM — fastest, XGBoost — most mature ecosystem, CatBoost — best for categorical features without preprocessing. 3) Time series → Prophet / NeuralProphet (Facebook) or statsforecast (Nixtla — state-of-the-art classical) or ML approach (LightGBM on feature-engineered time series — often beats specialised libraries). 4) NLP / text classification → Hugging Face Transformers (fine-tune small BERT/DeBERTa/RoBERTa) or scikit-learn TF-IDF + LogisticRegression if <10K docs (simpler baseline). 5) Computer Vision → PyTorch + torchvision (image classification / detection / segmentation). For production — vLLM / TGI. 7) Recommender systems → LightGBM / XGBoost on feature-engineered users × items (the 2026 workhorse), implicit / LightFM (collaborative filtering classics), PyTorch deep recsys (NeuMF / Two-Tower / Transformer4Rec). 8) Causal inference / uplift modelling → DoWhy / EconML / CausalML. 9) Bayesian modelling (when uncertainty quantification matters — medical / finance) → PyMC / Stan / NumPyro. 10) Anomaly detection → PyOD + Isolation Forest + LightGBM-based approaches. Default 2026: 80% of tabular tasks are solved by LightGBM / XGBoost / CatBoost; deep learning — for CV / NLP / recsys.

Yes, 26% of DS jobs are full-remote or hybrid. DS work — Jupyter + cloud data warehouse (Snowflake / BigQuery / Databricks / Redshift) + dashboards. Outsourcers — almost always remote on US projects. Russian product companies — hybrid or remote after probation. Russian banks — hybrid/office due to data security compliance. International tech companies — full-remote standard. Relocant hubs: Poland (Warsaw/Krakow — DS-friendly), Germany (Berlin / Munich), Canada (Toronto / Vancouver), Serbia, Georgia, UAE. English for international DS remote — must (premium +25-40%, and DS roles require stakeholder communication — English audible).

Product Data Scientist — a DS with deep product focus + ownership of growth experiments + retention modelling + feature-impact analysis. Works closely with PM and Engineering. Stack is the same, but the focus: A/B-test mastery (multi-variant experiments, sequential testing, multiple testing correction, power analysis), growth metrics (DAU/MAU/retention curves/funnel analysis), causal inference for observational data. Pay comparable to Senior DS (median), but at US product startups Product DS often exceeds general DS Senior thanks to business-impact ownership. General Data Scientist — can be in any domain (banking risk, retail demand, manufacturing quality). Research DS (or just "DS" in a research context) — focus on novel modelling approaches + papers (often PhD track). See also Research Engineer / Scientist. Career choice: Product DS if product impact is interesting and you enjoy stakeholder communication, general DS for domain flexibility, Research DS for new methodology research. All three are legitimate Senior+ paths.

Roadmap: 1) SQL mastery — must-have (window functions / CTEs / JOIN strategy / query optimisation). Book "SQL for Data Scientists" Tanimura. Course "Mode SQL Tutorial" (free). 2) Python for DS: pandas + NumPy mastery, not just syntax but performance patterns (vectorisation,.apply pitfalls). Book "Python for Data Analysis" McKinney. 3) Statistics: descriptive + inferential + hypothesis testing + ANOVA + linear regression deep. Course "Statistical Learning" Hastie / Tibshirani (StanfordOnline — free). Book "Practical Statistics for Data Scientists" Bruce. 4) Classical ML: scikit-learn mastery (pipelines / cross-val / metrics trade-offs / calibration), XGBoost / LightGBM / CatBoost (gradient boosting workhorses). Andrew Ng "Machine Learning Specialization" (Coursera) + "Introduction to Statistical Learning" Hastie (ISL — free PDF). 5) A/B testing mastery — power analysis, sample size, sequential testing, multiple testing correction, CUPED variance reduction. Books: "Trustworthy Online Controlled Experiments" Kohavi (must-read), "Statistical Methods in Online A/B Testing" Georgi. 6) Visualisation & storytelling — matplotlib + seaborn + plotly for notebooks + Tableau / Looker / Power BI / Apache Superset for stakeholders. Books: "Storytelling with Data" Knaflic, "The Visual Display of Quantitative Information" Tufte (classic). 7) Domain specialisation — pick one of fintech / healthcare / retail / telco / e-commerce and study domain-specific challenges. 8) Deep learning bonus: PyTorch basics for CV / NLP DS tasks. 9) Causal inference (Senior bonus): "The Effect" Huntington-Klein (free online textbook), DoWhy / EconML libraries. 10) Pet project: end-to-end DS project on Kaggle (data exploration → feature engineering → model selection → calibration → business presentation) + 1 A/B test simulation project. International (EN): DeepLearning.AI (Andrew Ng), fast.ai, Coursera IBM Data Science Professional. Analyst → Junior DS — 6-12 months. Backend Middle → Junior DS — 8-14 months (math + stats gap).

2377 active open DS positions. Geography: Russia / Poland / remote. The real market is broader thanks to a huge international remote segment. Time to close a Senior DS role — 4-10 weeks (faster than ML Engineer thanks to broader entry pool).

A Senior DS owns the full cycle from business problem to production model + technical leadership. SQL mastery: window functions / CTEs / complex JOINs / query optimisation / reading execution plans. Python for DS: pandas / NumPy / Polars performance patterns, vectorisation, profile-driven optimisation. Statistics mastery: hypothesis testing depth, Bayesian thinking, causal inference (DoWhy / EconML), Bayesian modelling basics (PyMC). Classical ML mastery: deep feature engineering, model selection rationale, calibration, business-aligned metrics, gradient boosting hyperparameter mastery (XGBoost / LightGBM / CatBoost). A/B testing mastery: power analysis, sequential testing, multiple testing correction, CUPED variance reduction, network effects handling, novelty/primacy effects, switchback experiments for marketplaces, geo-experiments. "Trustworthy Online Controlled Experiments" Kohavi — the desk book. Deep Learning (bonus): PyTorch for CV / NLP DS tasks. LLM-aware DS: prompt engineering, RAG basics, LLM-as-judge for evaluation, LLM applications in DS workflow (data exploration / synthetic data / labelling assistance). Visualisation & communication: storytelling-with-data mastery — convert results into business decisions for non-technical stakeholders. Cloud DS: one of AWS SageMaker / GCP Vertex AI / Databricks / Snowflake deeply. Soft: stakeholder management (product / business / engineering), technical writing (analysis docs, post-experiment reports, decision memos), mentoring Middle DS, hiring (interviewing DS candidates), executive communication. Domain expertise: deep understanding of one-two domains (fintech / healthcare / retail / telco / e-commerce) — the main premium driver for Senior+. English for Senior+ MUST — DS roles are intensely stakeholder-facing, and most materials / papers / community are in English.