A practical learning path through machine learning, deep learning, LLMs, and AI agents — with resources to study and ideas to build.
flowchart LR
A["Python + math"] --> B["Machine learning"] --> C["Deep learning"]
C --> D["LLMs + RAG"] --> E["AI agents"] --> F["Production"]
| Stage | Learn | Build |
|---|---|---|
| 01 · Foundations | Python, NumPy, linear algebra, probability | Explore and visualize a dataset |
| 02 · Machine learning | Regression, classification, clustering, evaluation | Train a classifier and explain its errors |
| 03 · Deep learning | Neural networks, backpropagation, PyTorch, transformers | Build an image or text classifier |
| 04 · LLM engineering | Embeddings, retrieval, RAG, structured outputs | Create document Q&A with citations |
| 05 · Agents | Tool use, state, MCP, human oversight, evaluation | Build and test a tool-using assistant |
| 06 · Production | APIs, monitoring, security, cost, reliability | Deploy and measure an AI application |
The stages are a guide, not a required sequence. Build small projects while you learn.
| Area | Resource | Why start here |
|---|---|---|
| ML | Google Machine Learning Crash Course | Applied fundamentals and exercises |
| Deep learning | Dive into Deep Learning | Explanations with executable code |
| Computer vision | Stanford CS231n | Vision models and deep learning |
| NLP | Stanford CS224N | Language models and NLP |
| LLMs | Hugging Face LLM Course | Transformers and hands-on tools |
| Agents | Building Effective Agents | Clear agent patterns and tradeoffs |
| MCP | Model Context Protocol documentation | Tools, resources, and integrations |
For deeper reading, see the ReAct and Toolformer papers. Resource links lead to their original publishers whenever possible.
The code examples above are present in the repository. The ideas below are future exercises, not completed projects.
Project ideas, from beginner to advanced
- Linear regression from scratch
- Spam classifier with error analysis
- Neural network trained on a small dataset
- Semantic search over personal notes
- Document Q&A with retrieval and citations
- Tool-using research assistant with task evaluations
- AI service with monitoring and cost tracking
For each project, document the dataset, baseline, evaluation method, limitations, and what you learned.
Found a better course, a broken link, or a useful project idea? Open an issue or send a pull request. Include the original source and a short note on why it belongs here.
This project grows from Shengjia Yan's ml-road resource collection. The original MIT notice is retained in LICENSE. Courses, papers, books, and other resources belong to their respective creators.
Built and curated by Ilgiz Satkynov · Learn → Build → Evaluate → Improve