G3 - search

keyword search is efficient when searching for exact matches but the pure lexical matching might struggle if the vocabulary differs
vector search intelligently returns semantically related information but certain keywords such as IDs, model numbers, etc lack rich semantic meaning, and it may even return another one, eg: $A B C 123$ may be considered close enough to $B C D 234$ , where they may be IDs
hybrid search combines lexical and vector retrieval signals
it does not simply concatenate everything without ranking
he system uses result fusion to combine/rank signals from the different retrieval methods.
retrieved candidates can be ranked semantically to improve ordering
retrieval recall measures the fraction of relevant chunks that were retrieved

R e c a l l @ k = \frac{r e t r i e v e d}{r e l e v a n t}

retrieval precision measures the fraction of retrieved chunks that are actually relevant

P r e c i s i o n @ n = \frac{r e l e v a n t}{n}

where, $n$ is the number of retrieved chunks

$k$ controls the recall-precision trade-off
reranking helps increase the precision
metadata filtering complements retrieval, eg: finding the information for a particular model
query rewriting is when the app transforms the user query into something more retrieval-friendly before searching
it must preserve the actual user intent