Structured learning path

AI Mastery Roadmap

A structured path from foundations to advanced AI systems. No prior background required.

~700 hourstotal

~12 monthsat 1–2 hrs/day

Beginner → Expertno prior background

5 phasesfoundation to mastery

Detailed phases

Phase 04–6 weeks · ~40–50 hours

Foundation

Get from zero to running Python and writing scripts. Variables, loops, functions, and a clear idea of what machine learning is.

Install Python 3, VS Code, and set up your environment
Variables, types, conditionals, loops, functions
Lists, dicts, and basic data structures
Small projects (calculator, text game, data parsing)
Intro: supervised vs unsupervised, training vs inference

Week-by-week breakdown

Week 1: Setup and variables • Install Python 3.11+ and VS Code • Variables, types (int, float, str, bool), basic operations • Input/output, f-strings • Practice: 5–10 small exercises daily Week 2: Conditionals and loops • if/elif/else, boolean logic • for loops (range, iterating lists) • while loops, break/continue • Practice: FizzBuzz, number guessing game Week 3: Functions and lists • Defining functions with def, parameters, return • Lists, indexing, slicing, list methods • Loops over lists, list comprehensions • Practice: Write 3–5 reusable functions Weeks 4–6: Consolidation • Small projects: calculator, text adventure, CSV parser • Dictionaries and basic file I/O • "What is ML?" — supervised vs unsupervised, training vs inference • Optional: NumPy basics (arrays, shape, indexing)

Project ideas

Simple calculatorText-based adventure gameCSV data analyzerCLI todo app

Core resources

→Python 3 official documentation
→Python Tutorial (docs.python.org)
→Automate the Boring Stuff (free online)

Expected outcome

You can run Python, write scripts with variables/loops/functions, use basic data structures, and explain what machine learning is in one sentence.

Phase 1Months 1–3 · ~130 hours

Math + Machine Learning

Linear algebra, probability, and calculus foundations. Then ML from scratch: linear regression, logistic regression, basic optimization. No black boxes.

Linear algebra: vectors, matrices, dot products, matrix multiplication
Probability: distributions, Bayes, expectation, variance
Calculus: derivatives, gradients, chain rule
Andrew Ng Machine Learning (Coursera) or equivalent
Implement linear regression and logistic regression from scratch
Gradient descent and normal equation

Week-by-week breakdown

Math (~25% of time, ~8 weeks) • Linear algebra: MML Ch. 2–4 — vectors, matrices, eigenvalues • Probability: basics, Bayes theorem, common distributions • Calculus: partial derivatives, gradients, chain rule • Derive key results on paper weekly ML (~75% of time, ~12 weeks) • Weeks 1–4: Linear regression — gradient descent, normal equation, feature scaling • Weeks 5–8: Logistic regression, classification metrics, regularization • Weeks 9–12: Optimization basics, bias-variance, model selection • One "from scratch" implementation per algorithm (no sklearn for core algos)

Project ideas

Predict house prices from scratchSpam classifier with logistic regressionImplement k-NN from scratchExploratory data analysis on a real dataset

Core resources

→Mathematics for Machine Learning (MML) Ch. 2–4 — free online
→Andrew Ng's Machine Learning (Coursera) — foundational
→3Blue1Brown — linear algebra and calculus intuition
→StatQuest (YouTube) — probability and stats

Expected outcome

You can derive and implement linear and logistic regression by hand; you understand gradients, optimization, and basic model evaluation.

Phase 2Months 4–6 · ~130 hours

Deep Learning

Neural networks, backpropagation, CNNs, attention, and transformers. Build implementation skills, use PyTorch/TensorFlow, and start reading papers.

Neural nets: forward pass, backpropagation, activation functions
CNNs: convolutions, pooling, ResNet-style architectures
Attention mechanism and transformer architecture
PyTorch or TensorFlow — build and train models
Paper reading: one paper per 2 weeks, implement key ideas
Stanford CS231n and DeepLearning.AI specialization

Week-by-week breakdown

Months 4–5: Neural networks and CNNs • Build a neural net from scratch (NumPy) for intuition • PyTorch/TensorFlow basics: tensors, autograd, nn.Module • MLPs, CNNs, batch norm, dropout • Image classification on CIFAR-10, MNIST Month 6: Attention and transformers • Attention mechanism, self-attention • Transformer architecture (Vaswani et al.) • BERT/GPT-style models — usage and fine-tuning • Start reading 1 paper per 2 weeks; take notes, implement snippets

Project ideas

Image classifier on custom datasetText classification with transformersReplicate a simple paper (e.g., ViT, BERT-base)Build a small GPT-style language model

Core resources

→Stanford CS231n (CNNs, vision) — free, excellent
→DeepLearning.AI specialization — structured path
→Fast.ai Practical Deep Learning — top-down approach
→The Illustrated Transformer (blog) — attention intuition

Expected outcome

You can implement and train neural networks in PyTorch/TensorFlow, understand CNNs and transformers, and have read and summarized 4–6 key papers.

Phase 3Months 7–9 · ~130 hours

Specialization & Systems

Deep dive into one track (NLP, CV, or RL). Add learning theory, deployment basics, and build a substantial project with clear documentation.

One track: NLP, Computer Vision, or Reinforcement Learning
Learning theory: bias-variance, overfitting, generalization, regularization
Systems: model deployment, FastAPI, basic MLOps
Paper reading: 2–3 papers per month in your track
Substantial project: replication or novel implementation
Write a clear technical report (5–10 pages)

Week-by-week breakdown

Specialization (choose one) • NLP: Transformers, fine-tuning, RAG, LLM applications • CV: Object detection, segmentation, video understanding • RL: MDPs, policy gradient, DQN, environments (Gym) Learning theory (parallel) • Bias-variance tradeoff, VC dimension (intuition) • Regularization, dropout, data augmentation • Cross-validation, hyperparameter tuning Systems (parallel) • Model serving with FastAPI • Docker basics, basic CI/CD • Logging, monitoring, experiment tracking Project (ongoing) • Choose replication or novel idea by Week 4 • Implement, experiment, iterate • Write up: problem, method, results, lessons

Project ideas

Fine-tune an LLM for a specific taskBuild an object detection pipelineTrain an RL agent (e.g., CartPole, Atari)Deploy a model as an API with monitoring

Core resources

→Track-specific: Hugging Face NLP, PyTorch CV, Spinning Up RL
→Paper: Attention Is All You Need, BERT, ResNet, DQN
→FastAPI documentation, Docker docs

Expected outcome

You have depth in one area, a written project report, deployment experience, and strong foundations for advanced work.

Phase 4Months 10–12 · ~130 hours

Advanced Mastery

Paper writing, rigorous replications, and advanced topics. Consolidate knowledge and produce work that demonstrates mastery. Capstone project.

Rigorous replication of 1–2 key papers in your specialization
Paper writing: clear problem, method, experiments, analysis
Consolidation: review full curriculum, fill gaps
Capstone: one major project or replication
Documentation, code quality, reproducibility
Optional: contribute to OSS, write blog posts

Week-by-week breakdown

Months 10–11: Capstone and replications • Select 1–2 papers to replicate with full rigor • Implement, debug, document, compare to reported results • Write a technical report (10–15 pages) or blog series • Seek feedback: mentors, study groups, online communities Month 12: Consolidation and polish • Review entire curriculum, re-derive key results • One "project-only" month: no new theory • Polish code, documentation, portfolio • Every 6th week: light review, no new content

Project ideas

Full replication of a seminal paperNovel extension or ablation studyOpen-source contribution to ML libraryTechnical blog series on a topic you mastered

Core resources

→Key papers in your specialization
→Reproducibility guidelines (e.g., Papers with Code)
→Technical writing: clear, concise, structured

Expected outcome

You have completed a structured path from foundations to advanced AI systems. You can demonstrate mastery through a capstone project, replications, and clear technical writing.

How to use this roadmap

1–2 hrs/day: Follow the phase structure. On busy days, do at least 30 min math + 30 min code.
Don't skip Phase 0. If you're new, start there. It gets you from zero to running Python.
Depth over breadth. Complete each phase before jumping ahead. Prerequisites matter.
Every 6th week: Light review — no new content. Re-read notes, re-derive key results.