Vector Search, Without the Maths

The idea in one sentence

Text is turned into a list of numbers positioned so that similar meanings sit close together. Searching becomes finding the nearest neighbours to your question, rather than matching its letters.

Why keyword search survives

Nobody wants a semantic match for a product code, an invoice number or an exact name. Precise queries deserve precise answers, and embeddings are confidently approximate.

Hybrid is the real answer

Run both, then merge the rankings. Keyword search anchors the exact terms; vectors catch the phrasing you did not anticipate. The merge is the part worth tuning, and it is mostly product judgement.

Chunking decides quality

How you split documents matters more than which model you embed with. Chunks that break mid-argument return confident, useless answers, and no amount of prompt work repairs a bad chunk.