I'm unhappy with Elasticsearch because hybrid search is too clunky

That’s a pretty common feeling. Elasticsearch can do hybrid search, but it often feels like you’re stitching together multiple systems rather than getting a clean, first-class search experience.

If “clunky” is what you’re hitting, the usual pain points are:

• Hard to tune lexical + vector ranking together
• Too much query plumbing for simple “search by meaning and keywords”
• Relevance is fiddly unless you spend a lot of time on boosts, rescoring, and evaluation
• Operational complexity: indexing, mappings, analyzers, ANN settings, and ranking logic all pile up

A few alternatives depend on what you want:

If you want simpler hybrid search

• OpenSearch: similar ecosystem, a bit more flexible in some hybrid setups, but still Elasticsearch-like complexity.
• Vespa: very strong for hybrid retrieval/ranking, but more of a platform than a plug-and-play search box.
• Typesense / Meilisearch: much simpler, but not as strong for advanced vector + keyword hybrid search.
• Postgres + pgvector + full-text search: surprisingly nice if your scale is moderate and you want fewer moving parts.

If your main issue is relevance quality

• Use a dedicated reranker after initial retrieval
• Or consider Vespa / managed vector search products that support multi-stage ranking more naturally

If your main issue is developer experience

A cleaner stack is often:

1. BM25 / keyword retrieval
2. vector retrieval
3. reranking
4. fusion logic in application code

That can be easier to reason about than forcing everything into one Elasticsearch query.

If you want, I can help in one of three ways:

1. Recommend a better replacement based on your scale/use case
2. Show a cleaner hybrid search architecture
3. Help simplify your current Elasticsearch setup so it’s less painful