Open any "how to become a data scientist" article and you'll find a list of 40 skills, 15 tools, 8 courses, and a timeline of "6–12 months." The list is real. The timeline is a lie told by people who want you to feel good about clicking enroll.
This roadmap is different. It's built for two types of people: someone in Hyderabad who wants to break into data from an IT background, and someone in Nairobi who has a non-technical degree and wants to understand where to start from zero. The path is the same. The local context differs. Both are accounted for here.
The honest version of this roadmap has three phases, not fifteen. Each phase ends when you can do something — not when you've watched enough videos. Here it is.
Where Most Roadmaps Go Wrong
The typical roadmap problem: it's written by someone who already knows data science, looking backwards. From that vantage point, everything feels important — statistics, linear algebra, SQL, Python, machine learning, deep learning, data engineering, cloud platforms. So they include everything.
The result is a roadmap that would take five years to follow properly, which means most people who try to follow it exactly either burn out in month three or spend years "learning" without ever building anything that could get them hired.
The real goal of a roadmap: Get to your first job as fast as possible. Your first job teaches you more than the next five courses combined. The roadmap's job is to get you in the door, not to make you comprehensively educated before you get there.
With that lens, here's what the path actually looks like.
The Actual Phases — What to Do and When
Foundation — Python and SQL
Python for data manipulation (pandas, NumPy, basic visualisation with matplotlib). SQL for querying and aggregating data. Nothing else. Do not start machine learning. Do not touch TensorFlow. Build three small projects using local datasets — something about Nairobi weather, Indian stock prices, Kenya county-level data, anything that interests you personally. The project is the point, not the syntax.
Analysis and Storytelling
Take a real dataset with a real business question. Walk through the full analysis cycle: load, clean, explore, visualise, conclude. Write up the finding in plain English. Present it to someone who doesn't know data. This phase is harder than phase one because the technical difficulty drops and the thinking difficulty rises. Most people skip it and go straight to machine learning. That is a mistake.
Machine Learning Basics
Supervised learning: classification and regression. Scikit-learn. Model evaluation (accuracy, precision, recall, ROC). Feature engineering. One end-to-end project: take a business problem, build a model, evaluate it honestly, present what it predicts and where it fails. You do not need to understand the full mathematics of every algorithm. You need to understand what each algorithm is for and whether your model is actually working.
Portfolio and Job Applications
Two to three solid projects, each solving a real business problem with a real dataset. For India: fintech, logistics, e-commerce data (Razorpay-style transaction analysis, Zepto-style demand forecasting). For Kenya: mobile money, agritech, retail data. GitHub profile that shows code quality, not just notebooks. LinkedIn that summarises findings in plain language. Start applying as soon as you have one complete project — do not wait for three.
The 4 Mistakes That Add 12 Months to Your Timeline
ML tutorials assume you can manipulate a dataframe, handle missing values, and write a for loop without looking anything up. If you can't, the ML tutorial becomes a copying exercise that teaches nothing.
Watching 40 hours of video is not the same as learning 40 hours worth of data science. The only thing that proves learning is something you built that wasn't there before.
These datasets are pre-cleaned, perfectly structured, and nothing like what you'll encounter at a job. Find messy, real-world data from your own context — government open data, M-Pesa public datasets, Indian census data. Learn to clean what isn't clean.
Ready doesn't exist. The job search is part of the learning. Applications tell you what skills employers actually care about right now in Hyderabad or Nairobi — not what a curriculum thought was important three years ago.
How to Know You're Ready for Each Next Step
Ready for Phase 2 when: you can load a CSV, clean nulls, filter by conditions, group and aggregate, and make a basic plot — without googling the syntax for each step.
Ready for Phase 3 when: you've taken a messy dataset, asked yourself a business question, and produced a finding you could explain out loud to a non-technical person in under three minutes.
Ready to apply when: you have one complete project with a clear question, a clear methodology, and a clear finding — and you can explain all three confidently.
Your Next Move
If you haven't started yet: install Python and pandas today. Find a dataset about something you actually care about. Try to answer one question about it. That's week one done.
If you're stuck in tutorial purgatory — watching videos, doing exercises, never building anything: stop. Close the course. Find a dataset. Ask a question. Build something broken. Fix it. That's the actual learning.
If you've been learning for more than 6 months and haven't applied to a job yet: apply this week. The feedback from real applications will teach you more than another month of courses.
Shorter than you think, harder than it sounds
The honest timeline for a first data analyst job: 5–8 months of focused, project-led work. Not watching videos. Not completing courses. Building things, breaking things, explaining findings, applying for jobs.
The people who get there fastest are not the ones who learn the most before they start applying. They're the ones who start applying earlier and learn from the rejection.
A structured path built around real data problems
The Data Science track at Ramayai is designed exactly around the phases above — Python first, analysis second, ML third, with projects grounded in East African and South Asian business contexts.
Explore the Data Science Track →