Before a neural network can process text, it must be broken down into numbers. Tokenization splits sentences into smaller units (tokens). Our model (MiniLM-L6) interprets these tokens to understand context, handling even complex words by breaking them into familiar sub-words.

We project the "meaning" of each sentence as a point in space. The distance between the blue planet (Sentence A) and the white orb (Sentence B) represents how similar they are. Closer objects = closer meaning. The beam color shifts from Red (Different) to Blue (Similar).

This heatmap visualizes the underlying math. The top rows show the raw numerical "thought" vectors for each sentence. The bottom row shows where they overlap. Bright Blue regions mean the sentences strongly agree on a specific feature; Red means they disagree.

Real-time logs of the inference process. We track how long it takes to encode text and calculate similarity, and we identify which specific abstract "dimensions" contribute most to the similarity score.