2. The roadmap: what to learn next

Data science stands on four legs: mathematics, statistics, programming and data engineering, and domain knowledge (health, climate, economics, agriculture…). Machine learning and AI build on all four.

                     ┌──────────────────────────────────────────────┐
                     │  Specialise: ML engineering · analytics ·    │
                     │  research · NLP · vision · geospatial · ...  │
                     └──────────────────────────────────────────────┘
                                          ▲
       ┌──────────────────┬───────────────┴─────────┬───────────────────┐
       │  Mathematics     │  Statistics             │ Programming & data │
       │  linear algebra  │  probability            │ Python, SQL, Git   │
       │  calculus        │  inference, regression  │ data engineering   │
       │  optimisation    │  experiments, causality │ software practice  │
       └──────────────────┴─────────────────────────┴────────────────────┘
                         ▲  + domain knowledge + communication + ethics  ▲
                         └──────────  your Grade 8–12 foundation  ─────────┘

Stage 1 (first 6–12 months after Grade 12)

Area	Learn	Why
Programming	Python properly: functions, data structures, files, modules, testing; Git and GitHub	everything else is built in code
Data	pandas fluently; SQL including window functions; data cleaning on real messy data	80% of real work
Maths	algebra review; functions; introduction to calculus (derivatives, rates of change)	for models and optimisation
Statistics	descriptive statistics, probability basics, distributions, sampling, intervals, tests, regression	judge evidence
Projects	2–3 complete projects with public write-ups	portfolio

Stage 2 (year 2)

Area	Learn
Linear algebra	vectors, matrices, matrix multiplication, solving systems, least squares, eigenvectors (the language of ML)
Calculus	derivatives, partial derivatives, gradients, chain rule (how models learn)
Probability	random variables, expectation and variance, common distributions, Bayes' rule, the law of large numbers and central limit theorem
Statistical modelling	multiple and logistic regression, model diagnostics, experimental design, causal thinking
Machine learning	the scikit-learn toolbox: trees, ensembles (random forests, gradient boosting), regularisation, cross-validation, pipelines, evaluation
Data engineering	relational database design, larger-than-memory data (DuckDB, Parquet), APIs, scheduling, cleaning pipelines
Visualisation	matplotlib/seaborn/plotly; dashboards; design principles

Stage 3 (years 3+): choose a direction

Direction	Focus	Typical roles
Data analysis / business intelligence	SQL, dashboards (Power BI, Tableau, Streamlit), statistics, communication	data analyst, MIS officer, M&E officer
Statistics and research	probability, inference, survey methods, causal inference, R or Python	statistician, research officer, survey specialist
Machine learning	ML theory, deep learning, MLOps	ML engineer, data scientist
Data engineering	databases, distributed data, cloud, pipelines	data engineer
Deep learning and AI	neural networks (PyTorch), NLP (including Nepali language technology), computer vision	AI engineer, researcher
Geospatial and earth data	GIS, remote sensing, climate and hazard data	GIS analyst, climate data scientist
Domain data science	health, agriculture, economics, education, development, journalism	epidemiologist, agricultural economist, data journalist

Nepal especially needs people who combine data skills with domain knowledge: public health, disaster risk, hydropower and climate, agriculture, Nepali-language technology, and good government statistics.

A 12-month self-study plan

About 8–10 hours a week alongside other study. Adjust freely.

Months	Main study	Project
1–2	Python properly (CS50P or Think Python); Git and GitHub	a cleaned, documented analysis of a Nepal dataset (RD-08 prices or RD-12 air)
3–4	Statistics (OpenIntro Statistics ch. 1–6, or Khan Academy Statistics); pandas deep practice	bootstrap and permutation analysis of a real question, written up as a blog post
5–6	SQL deep practice (SQLBolt, Kaggle "Advanced SQL"); data visualisation	dashboard of a public dataset with a clear question
7–8	Linear algebra (3Blue1Brown Essence of Linear Algebra, then MIT 18.06 or Linear Algebra Done Right); regression in depth	regression project with careful interpretation
9–10	Machine learning (Kaggle Learn ML courses; An Introduction to Statistical Learning)	end-to-end ML project with baseline, validation, fairness audit, model card
11–12	Choose a direction: deep learning (fast.ai), geospatial, analytics, or statistics	a capstone in that direction, with a client if possible

What not to worry about

• You don't need the newest tool or the most powerful computer. Fundamentals change slowly; tools change quickly.
• You don't need to be "a maths person". Persistence beats talent; mathematics becomes easier when you see why you need it.
• You don't need to learn everything. Depth in a few things plus the habit of learning is worth more than shallow knowledge of many.