When is it best to use the Query Decomposition skill?

It is most effective for queries involving comparisons ('vs'), relationships ('how does X affect Y'), or compound subjects where information is likely spread across different documents.

How does Reciprocal Rank Fusion (RRF) benefit search results?

RRF combines the ranked results from multiple independent concept searches into a single, optimized list without requiring manual parameter tuning, prioritizing documents that appear highly across multiple searches.

Does query decomposition increase search latency?

While it adds a processing step, this skill uses sub-millisecond heuristic checks for simple queries and parallelizes sub-query searches to keep the overall response time efficient.

What is query decomposition in a RAG context?

Query decomposition is the process of splitting a complex user question into smaller, independent sub-queries to ensure the retrieval system finds relevant documents for every component of the request.

Query Decomposition

Name: Query Decomposition
Author: yonatangross

byyonatangross

•

데이터 과학 및 ML

Decomposes complex, multi-concept queries into independent sub-topics to improve RAG retrieval accuracy and coverage.

The Query Decomposition skill optimizes Retrieval-Augmented Generation (RAG) by breaking down compound or multi-hop questions into distinct, searchable concepts. Using a combination of high-speed heuristic detection and LLM-driven extraction, it transforms complex user prompts into parallel search tasks. This ensures that documents covering different aspects of a query—such as comparisons or relationship-based questions—are all retrieved and then intelligently merged using Reciprocal Rank Fusion (RRF), leading to significantly higher recall and more accurate AI responses.

주요 기능

01Seamless integration with HyDE and vector search workflows

0269 GitHub stars

03Reciprocal Rank Fusion (RRF) for robust result consolidation

04Parallel retrieval execution to maintain low latency

05LLM-powered concept extraction for nuanced query partitioning

06Heuristic-based fast-path detection for common comparison operators

사용 사례

01Improving recall for compound questions spanning multiple domain topics

02Handling comparison queries like 'A vs B' in technical documentation search

03Executing multi-hop retrieval for complex architectural troubleshooting

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit query-decomposition

For use in Claude.ai and ChatGPT

Download Skill