AI Engineering Learning Resources

Two AI engineering learning resources are highlighted: a GitHub repository "ai-hands-on" and a Stanford University lecture. The repo provides Jupyter notebooks covering math, PyTorch basics, Transformer, and large model retrieval, guiding learners to implement neurons, layers, and attention mechanisms from scratch. It also includes practical tasks like image text recognition and book recommendations. The Stanford course (2 hours 34 minutes) covers tokenization, BPE, Transformer, pretraining, RLHF, DPO, and token-by-token generation, explaining how models like ChatGPT and Claude are built. Both resources emphasize self-implementation and understanding underlying mechanisms, suitable for beginners and those already developing AI agents.
Key insights
The hands-on repository provides a comprehensive bottom-up approach to AI engineering
The "ai-hands-on" GitHub repo covers mathematics, PyTorch fundamentals, Transformer, and large model retrieval entirely through Jupyter notebooks. Learners implement neurons, network layers, and attention mechanisms from scratch, and apply the knowledge to practical tasks like image text recognition. It also recommends classic books, forming a complete self-study path for understanding how AI systems run from the ground up.
The Stanford course condenses the full LLM pipeline into a single accessible lecture
The 2-hour-34-minute Stanford course walks through tokenization, BPE, Transformer, pretraining, RLHF, DPO, and token-by-token generation, breaking down the construction of ChatGPT and Claude. It is recommended for both newcomers to AI and those already building agents, offering a clear and complete overview of modern large language model development.
Both resources fill a gap in practical AI education by emphasizing implementation over theory
Many online tutorials skip the hands-on implementation layer, focusing only on theory or high-level frameworks. The highlighted repo and Stanford course directly address this gap by requiring learners to code core components from scratch and explain the entire LLM pipeline in a practical, applied manner.