本文へスキップ
TakoAPI
スキル一覧に戻る

rag-architect

alirezarezvani 作

Coding Agents & IDEs
89ダウンロード数はclawskills.shが公開している値です
1スター
0閲覧数
0件のいいねサインインするとこのスキルにいいねできます
—まだ評価がありません
あなたの評価

概要

The RAG (Retrieval-Augmented Generation) Architect skill provides comprehensive tools and knowledge for designing.

説明

このスキルでできること

RAG Architect covers the full lifecycle of retrieval-augmented generation pipelines, from document chunking and embedding selection to vector database setup, retrieval strategy tuning, and production deployment. It provides concrete guidance on tradeoffs between chunking methods, embedding models, and retrieval approaches including hybrid search and reranking.

Covers the entire RAG stack in one place, including tradeoffs that are typically scattered across separate documentation for chunking libraries, vector databases, and evaluation frameworks.

使いどころ

  • Building a Q&A chatbot over internal company documentation
  • Choosing a vector database for a production deployment
  • Debugging low retrieval precision in an existing RAG system
  • Setting up RAGAS evaluation for a knowledge base search feature
  • Reducing embedding API costs for a large document corpus

ワークフロー例

入力

User asks: How should I structure a RAG pipeline for a 10,000-page technical manual?

エージェント
  1. 1

    Analyzes document characteristics: structured technical content with headings and tables

  2. 2

    Recommends semantic or paragraph-based chunking with 10-20% overlap to preserve document structure

  3. 3

    Suggests a balanced embedding model such as all-mpnet-base-v2 for general technical content

  4. 4

    Recommends pgvector or Qdrant based on existing infrastructure and scale requirements

  5. 5

    Outlines RAGAS evaluation setup with faithfulness above 90% and context relevance above 0.8 as targets

出力

A complete pipeline recommendation with specific chunking strategy, embedding model, vector database, and evaluation metrics tailored to the document type and scale

インストール

clawhub install alirezarezvani/rag-architect

CLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。

リンク