What problem does it solve? Building a working AI prototype is easy, but making it reliable, scalable, and cost-effective in production is hard. This Skill provides the operational patterns for RAG pipelines, embeddings, vector databases, prompt engineering, semantic caching, cost estimation, and quality evaluation of LLM applications. ## Core Features & Use Cases - RAG Pipeline Implementation: Chunking, embedding, indexing, and querying documents with Chroma, pgvector, Pinecone, Weaviate, or Qdrant, including relevance filtering by distance threshold. - Prompt Engineering & Cost Control: Structured system prompt design, chain-of-thought analysis, semantic caching to avoid duplicate LLM calls, and monthly cost estimation across Claude models. - Quality Evaluation Framework: LLM-as-judge evals scoring responses against criteria like factual accuracy, relevance, and clarity. - Use Case: You need to add document search to your product. Use this Skill to set up a pgvector table with cosine similarity indexes, index your documents with overlapping chunks, and build a query pipeline that injects the top-k relevant chunks into the LLM context. ## Quick Start Ask the assistant to set up a complete RAG pipeline with pgvector for indexing your project documentation and answering questions from it.
