Generate multiple query variants and fuse results with Reciprocal Rank Fusion.
npx skills add https://github.com/juanre/llmemory --skill multi-query
Please help me install this Agent Skill. Skill: multi-query Source: https://github.com/juanre/llmemory/tree/main/skills/multi-query Command: npx skills add https://github.com/juanre/llmemory --skill multi-query
Multi-query expansion improves search recall by generating multiple variants of a user query, querying each variant, and fusing results using Reciprocal Rank Fusion (RRF) to yield a unified, deduplicated result set.
Enable query expansion for a hybrid search to generate multiple variants and fuse results with RRF. Example: memory.search(owner_id='workspace-1', query_text='improve customer retention', search_type=SearchType.HYBRID, query_expansion=True, max_query_variants=3, limit=10).
High-intent search queries and answers about installing and using this skill.
Multi-query expansion improves vector search recall by generating multiple query variants, executing searches for each, and fusing the deduplicated results using Reciprocal Rank Fusion (RRF) to boost consistently relevant chunks.
You can generate query variants without an LLM by using heuristic mode, which automatically creates keyword variants through stopword removal, OR logic, and exact phrase matching to widen search recall.
Reciprocal Rank Fusion (RRF) is a fusion strategy that combines and deduplicates result sets from multiple query variants, boosting chunks that appear across different variant searches to yield a unified list.
Yes, you can enable optional LLM-based expansion to generate semantically diverse query variants, providing an alternative to the default heuristic keyword generation for handling ambiguous queries.
You can apply per-query overrides and performance tuning by configuring parameters such as search_type, query_expansion, max_query_variants, and limit directly within your hybrid search request.
Use multi-query expansion for information retrieval when facing ambiguous or under-specified queries, as generating multiple variants and fusing results with RRF ensures efficient, accurate recall across hybrid vector and text scenarios.
RamVegirajuCreate and query Databricks vector search endpoints and indexes.
latestaiagentsCombine dense vector embeddings with sparse BM25 retrieval for RAG systems.
jkitchinTranslate user queries into vector and full-text searches in litdb.
hltcoeDeploy and query retrieval pipelines as a single RoutIR service.
