This article explains how text is converted into numerical vectors for machine learning, starting with foundational concepts like TF-IDF. It begins by questioning how machines understand language and respond to queries, emphasizing that models process numerical representations of data. The author advocates building from basics before advancing to complex topics like embeddings or transformers.

TF-IDF is introduced as a classic method for text representation, combining Term Frequency (how often a word appears in a document) and Inverse Document Frequency (how rare a word is across documents). The calculation TF-IDF = TF × IDF is detailed. A simple dataset of three food-related sentences is used to demonstrate tokenization, vocabulary building, and step-by-step computation.

The piece covers implementation in Python, vector visualization via PCA, and application in text classification. It also addresses handling unseen words and limitations of TF-IDF, positioning it as a stepping stone toward modern embeddings. The tone is educational and exploratory, aimed at learners in AI/ML seeking intuitive understanding.