What is Data Science ML/AI?
Data Science ML/AI turns raw data into decisions, predictions, and automated systems. You’ll collect and clean data, analyze patterns, build statistical and machine learning models, and communicate results to guide product and business choices. Typical problems include demand forecasting, customer segmentation, churn prediction, fraud detection, A/B testing, recommendation systems, time-series forecasting, and NLP tasks like sentiment or topic analysis.
Examples of business questions you’ll help answer
- Which customers are likely to churn next month, and why?
- What price maximizes revenue without hurting retention?
- Which users should see which content or product next?
- How can we detect fraud in near real-time?
- Which features in our app actually move core metrics?
Who this is for
- People who enjoy puzzles, patterns, and asking “why.”
- Comfortable with structured thinking, basic math, and some coding—or willing to learn.
- Curious about how products and businesses work and want to influence decisions with evidence.
Prerequisites
- Math basics: functions, algebra, and comfort with percentages and ratios.
- Statistics basics: mean/median, variance, sampling, probability, A/B testing ideas.
- Basic coding: beginner Python or R plus SQL fundamentals (SELECT, WHERE, JOIN, GROUP BY).
- Mindset: curiosity, patience with messy data, willingness to iterate.
Learning path
- Foundations: SQL, Python/R, descriptive stats, data cleaning, visualization.
- Analysis: experimentation (A/B testing), regression, classification basics, feature engineering.
- Machine Learning: model training/validation, cross-validation, regularization, tree-based models, basic NLP/time-series.
- Production awareness: version control (Git), notebooks to scripts, simple APIs, monitoring, documentation.
- Specialization (optional): recommender systems, NLP, computer vision, causal inference, time-series, ML Ops.
Careers inside this direction
Data Scientist
Builds and evaluates models, runs experiments, and turns data into decisions. Balances analysis, modeling, and stakeholder communication.
- Best for: people who enjoy problem framing, modeling, and explaining results.
Where you can work
- Industries: tech, fintech, e-commerce, healthcare, gaming, media, logistics, SaaS, government, NGOs.
- Company types: startups (broad responsibilities), scaleups (fast experiments), enterprises (specialized roles), consultancies (varied clients).
- Common teams: product analytics, marketing analytics, risk/fraud, platform/ML, research/innovation.
Salary ranges by stage
Varies by country/company; treat as rough ranges.
- Junior: ~$50k–$90k USD
- Mid-level: ~$90k–$140k USD
- Senior/Lead: ~$140k–$220k+ USD
What shifts pay up or down?
- Location and cost of living
- Domain (e.g., fintech and ads often pay more)
- Impact on revenue and responsibility scope
- Production ML experience and mentorship leadership
Growth map
- Level 1: Clean data, answer defined questions, create clear charts, basic SQL + Python, simple regression/classification.
- Level 2: Frame problems, design A/B tests, build robust models (trees/ensembles), communicate trade-offs, document assumptions.
- Level 3: Own a metric area, productionize models with engineers, set monitoring, mentor others, drive roadmap with stakeholders.
- Level 4: Cross-team strategy, model and platform standards, experimentation culture, hiring and capability building.
Signals you’re ready for the next level
- Consistently reproducible work with versioning and tests
- Clear stakeholder narratives that change decisions
- Modeling choices tied to business constraints and risk
- Post-deployment monitoring and iteration
Tools & stack overview
- Languages: Python (pandas, scikit-learn), R (tidyverse, caret)
- Data: SQL (PostgreSQL, MySQL, BigQuery, Snowflake), files (CSV/Parquet)
- Exploration: Jupyter/VS Code notebooks, RStudio
- Visualization: matplotlib, seaborn, Plotly; ggplot2
- ML: scikit-learn, XGBoost, LightGBM; basics of PyTorch/TensorFlow if going deep into ML
- Productivity: Git, virtual environments, Makefiles or simple scripts
- Ops (intro): APIs, Docker basics, monitoring metrics
Beginner roadmap (4–8 weeks)
Pick 6 weeks if you can; extend to 8 if needed. Keep sessions short and consistent.
Week 1: Data & SQL foundations
- Install a SQL environment (local or cloud sandbox). Practice SELECT, WHERE, ORDER BY.
- Learn JOINs and GROUP BY with sample datasets.
- Mini task: Write a query to find top 5 products by revenue last month.
Week 2: Python/R and data cleaning
- Set up Python (pandas) or R (tidyverse). Load CSVs, inspect data types, handle missing values.
- Compute basic stats and create descriptive charts.
- Mini task: Clean a messy dataset and summarize 3 key insights.
Week 3: Stats for decisions
- Probability, distributions, confidence intervals, hypothesis testing.
- A/B testing: metrics, significance vs. power, pitfalls.
- Mini task: Simulate an A/B test and interpret results.
Week 4: Core ML
- Train/test split, cross-validation, evaluation metrics (accuracy, ROC AUC, RMSE).
- Baseline models: linear/logistic regression, decision trees, random forests/GBMs.
- Mini task: Build a simple classifier and explain the top features.
Week 5: Communication & reproducibility
- Turn your notebook into a clear report with assumptions and limitations.
- Use Git to version your project; add a README and environment file.
- Mini task: Create 2 stakeholder-friendly charts with short captions.
Week 6: Capstone & light deployment
- Choose a dataset aligned with a business question and deliver an end-to-end analysis or model.
- Package key steps as functions or a simple script; add basic tests (if comfortable).
- Optional: Expose a prediction function with a tiny API or CLI script.
If you have 2 extra weeks
- Week 7: Time-series or NLP basics; add a second model type.
- Week 8: Experiment design deep dive; learn monitoring and drift checks.
Common mistakes
- Jumping to complex models before understanding the problem and metric.
- Ignoring data quality and leakage; not creating a clean validation split.
- Evaluating with the wrong metric for the business goal.
- Overfitting to a benchmark dataset; not testing generalization.
- Unclear communication: sharing code, not decisions and trade-offs.
- No reproducibility: random results that others can’t run.
Mini project ideas
- Churn classifier: predict who cancels and propose retention actions.
- Demand forecast: predict weekly sales; compare naive vs. ML baseline.
- Recommender mini: item-to-item similarity using co-occurrence.
- NLP: classify support tickets by topic; surface top drivers of complaints.
- Experiment analysis: simulate an A/B test; present a go/no-go decision.
How to present your project
- 1 slide: problem, metric, constraints.
- 1 slide: data quality and key features.
- 1 slide: model results vs. baseline.
- 1 slide: business impact and next steps.
Quick fit test
Take the short fit test below to see how your interests align. Everyone can take it for free; only logged-in users get saved progress.
Next steps
- Pick a starting role and commit to the 6-week roadmap.
- Build 1–2 mini projects and present them clearly.
- Then open the Careers section on this page to choose your path and dive deeper.