Concepts

Learn how text is split into tokens, why subword tokenizers exist, and how tokenization affects LLM behavior and cost.

Understand what large language models are, how they predict the next token, and why scale matters.

Learn how neural networks learn patterns through layers, weights, and backpropagation.

Learn the core techniques for writing effective prompts: system messages, few-shot examples, and structured instructions.

Learn how RAG lets an LLM answer questions using relevant external documents fetched at query time.

Learn how embeddings turn text into vectors and enable semantic search by finding meaning-based similarity instead of keyword matches.

Learn what vector databases store, why nearest-neighbor search must be approximate at scale, and how ANN indexes (like HNSW and IVF) make retrieval fast.

Build a practical mental model for context limits and how to allocate tokens for better cost, speed, and answer quality.

Understand how token selection strategies control output quality, diversity, and consistency.

Learn how attention helps models decide what matters, from query-key-value math to multi-head behavior in modern transformers.

Learn why LoRA-style parameter-efficient tuning is the default in practice and how to choose robust fine-tuning recipes.

Understand how models call external code safely and reliably using structured outputs, validation, and execution boundaries.

Understand how Transformers use attention to process sequences in parallel and power modern LLMs.

Learn how to split documents into retrievable chunks, attach the right metadata, and index content so RAG retrieves the right context reliably.

Learn when to shape an LLM with prompts versus when to change its behavior with fine-tuning, and the trade-offs of each.

Learn why two-stage retrieval and keyword+vector fusion improve relevance in real-world RAG systems.

Trace how base models become assistants through supervised instruction tuning and preference optimization methods like RLHF and DPO.

Learn how LLM agents execute multi-step workflows with planning, tool loops, recovery logic, and safety boundaries.