Hybrid keyword and vector search

Two methods with complementary blind spots

Vector search finds meaning but misses exact tokens: an error code, a product SKU, a surname. Keyword search nails those and misses every paraphrase.

Running both and fusing the ranked lists gives you the union of their strengths. Reciprocal rank fusion is the standard approach and needs no score calibration between the two systems, which is what makes it robust.

Hybrid search is usually the single largest quality improvement available to a pipeline that started with vectors alone.