Awesome AI Roadmaps
Free AI and machine learning roadmaps, from Python and math to deep learning, LLMs, and AI safety. Each step links to a specific video, paper, or chapter in learning order. 25 tracks, 258 steps.
Read on the website or browse the tracks below.
Contents
| Track | Covers | Steps |
|---|---|---|
| Python for ML | NumPy, notebooks, pandas, plots. | 9 |
| Math for ML | Linear algebra, calculus, probability, optimization. | 10 |
| Learn NLP | Word vectors through BERT. | 14 |
| ML Basics | Regression, trees, SVM, clustering. | 12 |
| Deep Learning | Backprop, CNNs, RNNs, optimization. | 12 |
| Computer Vision | CNNs, detection, segmentation, ViT. | 12 |
| LLMs | Transformers, GPT family, scaling, RLHF. | 12 |
| From Scratch | Autograd, tokenizers, GPT, all hand-built. | 7 |
| Eval Harnesses | Benchmarks, contamination, SWE-bench, LLM-as-judge. | 8 |
| RAG | Embeddings, chunking, vector search, eval. | 8 |
| Agents & Tooling | Tool use, ReAct, memory, browser/computer-use. | 7 |
| AI Tools | Cursor, Claude Code, local models, MCP, playgrounds. | 6 |
| Prompting & Context | Prompt structure, chain-of-thought, context engineering. | 6 |
| Fine-Tuning | LoRA, QLoRA, DPO, data for adapters. | 6 |
| Speech & Audio | Spectrograms, CTC, wav2vec, Whisper, TTS. | 11 |
| Multimodal | CLIP, Flamingo, BLIP-2, LLaVA, ImageBind. | 11 |
| Generative Models | VAE, GAN, flows, diffusion, flow matching. | 12 |
| Reinforcement Learning | MDPs, Q-learning, DQN, policy gradients, PPO, SAC. | 11 |
| Graph ML | Node embeddings, GCN, GraphSAGE, GAT, GIN, TransE. | 11 |
| Recommender Systems | Collaborative filtering, matrix factorization, Wide & Deep, two-tower, SASRec. | 12 |
| Time Series | Stationarity, ARIMA, ETS, DeepAR, N-BEATS, TFT, Informer, PatchTST. | 13 |
| Causal ML | Potential outcomes, DAGs, propensity scores, double ML, LATE, causal forests. | 11 |
| Interpretability | Permutation importance, PDP, LIME, SHAP, Grad-CAM, TCAV, circuits. | 13 |
| Evals & Safety | Alignment, specification gaming, jailbreaks, red teaming, safety evals. | 12 |
| Data-Centric AI | Datasheets, labeling, weak supervision, label errors, synthetic data, filtering. | 12 |
Where to start
- New to machine learning: Python for ML → Math for ML → ML Basics → Deep Learning.
- Build with language models: LLMs → Prompting & Context → RAG → Eval Harnesses.
- Implement the models: start with Deep Learning, then follow From Scratch.
Check each track’s prerequisites. Work through its numbered steps in order; use the video, reading, or both. YouTube links are bold. An empty cell means that medium is not listed.
Why this exists
Each track puts concepts in learning order and links directly to lessons, papers, and chapters. Follow a track from the top, or use its prerequisites to find the preparation you need.
Resources are selected for free access. Paid courses, certificate walls, and DSA/interview material are out of scope. Links can change; report unavailable resources through the issue forms below.
Adding a resource
Use the issue forms to suggest a resource or report a broken link, paywall, or learning gap. See contributing.md for the selection criteria. Please use issues for resource suggestions rather than pull requests.