Here's the hybrid vector+lexical scoring trick nobody explains.

We're OpenSolr - Solr hosting and consulting. We're obsessed with search (probably too much).

When we added vector search to Solr, we hit a problem nobody talks about: combining scores.

Vector similarity: 0 to 1
Lexical (BM25/edismax): 0 to whatever

Naive sum = lexical always wins, even when semantically wrong.
Fix: normalized_lexical = lexical / (lexical + k)

Now we have:

Cross-lingual search (EN→RO)
Emoji search (🔥 finds fires, 🐕 finds dog products)
Semantic fallback (wine emoji finds champagne when no wine exists)
Full debug inspector on every search

Live demos you can try:

https://opensolr.com/search/dedeman?q=🔥wood (Romanian hardware)
https://opensolr.com/search/vector?q=🔥 (news)
https://opensolr.com/search/peilishop?q=winter+hat (fashion)

Click the debug button to see actual Solr params. We built it to be educational.
Solr 9.x has dense vector support. You don't need Pinecone.

If you're fighting relevance issues or want help with hybrid search, that's literally what we love doing.
Happy to give pointers.