Data & AI

Frequently asked questions

Do I need a CS or stats degree to break into data?

No, but you'll need to demonstrate the same skills the degree would have given you. Hiring managers look for fluent SQL, comfortable Python, a working grasp of statistics (sampling, hypothesis tests, confidence intervals), and one finished project that solved a real-looking question end-to-end. Degrees still help for research-engineer and applied-scientist roles at frontier labs. For analytics, analytics-engineering, and the majority of ML engineering roles, a portfolio plus the right take-home performance is what gets you to a final round.

Should I learn classical machine learning or jump straight to LLMs?

Learn classical ML first if you have any flexibility on timing. The fundamentals — train/test splits, overfitting, regularisation, evaluation metrics — transfer directly to LLM work and will save you from a class of mistakes that have nothing to do with prompts. Two months on supervised learning with scikit-learn, plus one Kaggle-style finished project, is a reasonable floor before pivoting to LLM engineering. If you're already a working engineer who just needs to ship a RAG feature this quarter, skip ahead — but circle back to the fundamentals within a year.

What's the difference between prompt engineering and LLM engineering?

Prompt engineering is the craft of getting reliable behaviour out of a single model call — wording, structure, examples, output format. LLM engineering is the broader discipline that wraps prompts inside a real system: retrieval, evals, observability, fallback chains, agent loops, and the production infrastructure around all of it. Most working roles are LLM engineering with a strong prompt-engineering subskill. Pure prompt-engineering courses are short and high-leverage; the LLM engineering tracks on this pillar are where the longer, system-level material lives.

How important is SQL for data and AI roles in 2026?

Still critical, and probably more important than the generative-AI hype cycle suggests. Every data role still pulls from a warehouse or operational database, and analytical SQL — window functions, CTEs, query plans — is the single skill that crosses analyst, analytics engineer, data engineer, and ML engineer roles uniformly. The Databases topic on this pillar covers it alongside the design and operations side. If you can only learn one thing this month and you're targeting a data job, make it SQL.

Will LLMs make traditional ML obsolete?

No. They've absorbed natural-language and some vision tasks, which is significant, but the majority of business-critical predictions still run on tabular data — fraud scoring, customer churn, demand forecasts, pricing, credit decisions — where gradient-boosted trees and similar classical methods remain faster, cheaper, more accurate, and more interpretable than any LLM. Job postings reflect this: classical-ML roles are not disappearing. What's changing is that more teams now also need someone who can integrate an LLM where the inputs are unstructured text, which is why the pillar covers both sides.

How long does it take to become job-ready in data?

Six to nine months of consistent study (an hour a day or two evenings a week) is enough for an analytics-engineer or junior-analyst role if you finish one substantial portfolio project on a real dataset. ML engineering is closer to 12–18 months because the math and the systems work both take time. LLM-engineering roles can come faster — three to six months — if you arrive already comfortable with Python and APIs, since most of the work is system design rather than new fundamentals.