Explanation of each term below-> LLM: predicts the next word, one at a time, based on everything before it. Scale that up and you get answers, code, and conversation from the same mechanism. Tokenization: text becomes tokens before the model reads it, each one a word or a fragment of one. The model never sees individual letters, only these chunks. Context window: how much the model can hold at once, measured in tokens. Anything past the edge does not exist to it. Embeddings: turning text into numbers that carry meaning. Similar ideas end up close together, so you can compare meaning mathematically. Vector database: storage built for those numbers, able to find the closest matches across millions of entries in milliseconds. It searches by meaning, not keywords. Cosine similarity: the math behind “closest”. It compares the direction of two embeddings rather than their size. RAG: the model does not know your documents, so you fetch the relevant ones first and hand them over with th...