Skill
RAG Implementation
Builds retrieval pipelines that ground answers in real documents.
Install
/plugin marketplace add wshobson/agents; /plugin install llm-application-dev
Triggers on
When an LLM needs to answer from a document set, or an existing RAG pipeline returns weak matches.
- Author
- wshobson
- License
- MIT
Repositoryhttps://github.com/wshobson/agents
Every stage of a retrieval-augmented pipeline gets a decision rather than a default: how documents are split, which embedding model fits the language and the budget, and which vector store to run — Pinecone, Weaviate, Qdrant, Chroma or pgvector, compared on what separates them in practice. Retrieval is treated as more than nearest-neighbour lookup, combining dense and keyword search and reranking the result before it reaches the model. A LangGraph example shows retrieval and generation as separate nodes so each can be measured on its own.
Comments(0)
Sign in to comment