rag-architect

Building a retrieval-augmented generation system involves chunking documents, generating embeddings, storing them in a vector database, and fetching relevant context to answer queries. This skill provides detailed instructions for constructing and deploying these systems end-to-end.

Improving retrieval accuracy in a RAG pipeline involves using hybrid retrieval methods that combine dense and sparse search, followed by reranking with cross-encoders. This skill provides methods to enhance relevance through these advanced retrieval techniques.

Yes, this skill supports ChromaDB and Pinecone for vector storage. It facilitates selecting suitable embedding models and managing these vector databases for efficient semantic similarity search.

Yes, you can use sentence_transformers for generating embeddings. This skill provides guidance on embedding choice and implements strategies for dividing large texts into meaningful parts for custom document search applications.

Evaluating retrieval fidelity and answer faithfulness requires specific metrics and frameworks to assess system robustness. This skill provides evaluation and monitoring tools to measure retrieval accuracy and response quality.

Production deployment patterns for retrieval-augmented generation include caching, streaming, fallback, cost control, and safety measures. This skill covers architecture patterns and pitfalls to avoid when deploying enterprise RAG systems.