No degree required to start. A practical, week-by-week plan that takes you from zero to job-ready in about six months — built around real projects, not just theory.
Watching tutorials without writing code is wasted time. Every concept below is paired with something you build. Ship small things often.
Who it suits: You enjoy math, experimentation, and building systems that learn from data.
Indicative 2026 starting ranges in India — actual pay varies widely by city, company, and skill. Salary is an outcome of skill and a strong project portfolio, not a guarantee.
Get comfortable with Python and just enough math.
Variables, loops, functions, lists & dicts, files, error handling.
NumPy arrays, pandas DataFrames, matplotlib charts. Clean a real CSV.
Vectors, matrices, dot products, gradients — the intuition, not proofs.
Mean, variance, distributions, Bayes, correlation vs causation.
Understand and apply classic ML with scikit-learn. (Andrew Ng's ML Specialization.)
Cost functions, gradient descent, train/test split. Start the House Price Predictor.
Logistic regression, accuracy, precision, recall, F1, confusion matrix.
Overfitting, bias-variance, regularization, cross-validation, scaling.
Decision trees, random forests, gradient boosting, feature importance.
KNN, SVM, Naive Bayes, and GridSearch hyperparameter tuning.
K-means clustering and PCA for dimensionality reduction.
Missing data, outliers, encoding, full sklearn pipelines.
Finish the predictor end-to-end with a clean GitHub README.
Build neural networks with PyTorch — from MNIST to Transformers.
Neurons, activations, loss, backpropagation, optimizers.
Tensors, autograd, your first trained network on MNIST digits.
Convolution & pooling. Start the Image Classifier.
Train on a real image dataset, plot learning curves.
Dropout, batch norm, augmentation, and transfer learning.
Embeddings, RNN/LSTM basics, text preprocessing.
Attention — the foundation of ChatGPT. Use Hugging Face.
Finish & deploy a demo on Hugging Face Spaces or Streamlit.
Ship models like an engineer — the part that turns "learner" into "hireable".
Git, clean project structure, virtual envs, modular code.
Serve a model with FastAPI, Docker basics, deploy to a free host.
Call an LLM API, build document Q&A with embeddings + vector search.
One polished end-to-end app. Portfolio, resume, start applying.
Projects beat certificates. Put each one on GitHub with a clear README — recruiters look at what you've built.
Regression with pandas & scikit-learn. Compare Linear Regression vs Random Forest.
Text features, logistic regression, precision/recall — pick spam, sentiment, or Titanic.
A CNN in PyTorch with transfer learning; deploy an upload-an-image demo.
Upload a PDF, ask questions. Embeddings + vector search + an LLM, in a Streamlit UI.
Model served via FastAPI, containerized with Docker, live on a free host.
Click any resource to open it — free and well-known first, then build something with it.
Universal (any field): roadmap.sh · freeCodeCamp · GitHub Student Pack · Coursera
Go to colab.research.google.com — free, in your browser, every library pre-installed. Zero setup.
Create a free account at github.com. This is where your projects live and where recruiters look.
Load the California Housing dataset, plot a few charts, and train your first model. That's it — you've started.
This roadmap is for AI/ML. PathAhead helps you choose the right path in the first place — with guidance built for Intermediate students.
This roadmap is a free educational resource. Timelines are a guide — everyone learns at a different pace. Resource names (Coursera, fast.ai, freeCodeCamp, 3Blue1Brown, Khan Academy, Hugging Face) are referenced for learning only and are not affiliated with PathAhead.