skills.rest
TheViziusGroup
TheViziusGroupOFFICIAL
Explore all skills by this author →
WebsiteGitHub

rag-and-agents

Designs and evaluates RAG pipelines and AI agent systems on Azure and open-source frameworks.

1|Updated Aug 10, 2026
Search Tags:#rag#vector database#ai agents#langgraph#chunking#azure ai search#retrieval evaluation
One-click install
npx skills add https://github.com/TheViziusGroup/vibe-engineering-skills --skill rag-and-agents-theviziusgroup
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rag-and-agents
Source: https://github.com/TheViziusGroup/vibe-engineering-skills/tree/main/plugins/ai-and-data/skills/rag-and-agents
Command: npx skills add https://github.com/TheViziusGroup/vibe-engineering-skills --skill rag-and-agents-theviziusgroup

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Building retrieval-augmented generation pipelines and AI agents involves dozens of interdependent decisions—parsing, chunking, embeddings, vector databases, reranking, evaluation, and agent orchestration—where wrong choices cause hallucinations, poor retrieval quality, and runaway costs. This Skill provides a production reference that guides those decisions with verified benchmarks and staged implementation guidance. ## Core Features & Use Cases - RAG Pipeline Design: Covers document parsing (LlamaParse, Docling, Azure Document Intelligence), chunking strategies (parent-child, contextual retrieval), embedding model selection, vector database comparison, hybrid search with reranking, and GraphRAG. - Agent System Architecture: Compares LangGraph, CrewAI, Microsoft Agent Framework, and Foundry Agent Service, with guidance on MCP, multi-agent patterns, memory tiers, sandboxed code execution, and production controls like iteration limits and budget caps. - Evaluation & Azure-Native Solutions: Details RAGAS metrics, golden QA dataset construction, Azure AI Search hybrid/semantic configuration, Foundry IQ, and a staged implementation roadmap. - Use Case: When your RAG chatbot returns irrelevant answers, use this Skill to diagnose whether the issue is chunking, retrieval, or reranking, then apply the appropriate fix such as contextual retrieval or hybrid search with a cross-encoder reranker. ## Quick Start Ask the AI to help design a RAG pipeline for your document corpus or to debug why your existing retrieval system returns irrelevant chunks.

Frequently Asked Questions about rag-and-agents

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I improve RAG retrieval quality?▼

Start with hybrid search combining BM25 and vector search fused via RRF, then add a cross-encoder reranker over the top 50-200 results. If misses persist, apply parent-child chunking or Anthropic's contextual retrieval, which reduces retrieval failures by up to 67% when combined with reranking.

What chunking strategy works best for RAG?▼

Parent-child chunking is the highest-ROI production pattern: embed small child chunks of 100-500 tokens for retrieval precision, then return the larger parent chunk of 500-2,000 tokens to the LLM for generation context. Size guidance varies by use case, from 512 tokens for FAQs to 2,048 for legal contracts.

Which vector database should I use for RAG?▼

Qdrant offers best-in-class filtered search, Weaviate has strong native hybrid search, and pgvector is sufficient under roughly 10 million vectors if you already run Postgres. On Azure, AI Search provides hybrid search with a semantic reranker, while Cosmos DB with DiskANN handles billions of vectors at low latency.

LangGraph vs CrewAI vs Microsoft Agent Framework for agents?▼

LangGraph is the production default for stateful, auditable workflows with checkpointing and human-in-the-loop interrupts. CrewAI suits fast role-based multi-agent prototyping, while Microsoft Agent Framework 1.0 is the open-source Azure-aligned successor to Semantic Kernel and AutoGen.

When should I use GraphRAG instead of standard vector RAG?▼

Use GraphRAG only for multi-hop or thematic queries across a large, relatively static corpus, where it beats vector RAG on comprehensiveness by 72-83%. Start with LazyGraphRAG, which costs roughly 0.1% of full GraphRAG indexing, and never deploy full GraphRAG on high-update or simple-factoid corpora.

Why does my RAG system hallucinate despite retrieval?▼

Hallucination persists when generation is not grounded: add an explicit anti-hallucination system prompt instructing the model to answer only from provided context, place the best material at the start or end of the context window to avoid lost-in-the-middle degradation, and measure faithfulness with RAGAS.

Related Skills

Explore More Software Engineering Skills→
vecearvecear

embedding-pipeline-builder

Build document embedding and retrieval pipelines for scalable semantic search.

Community
Advanced
latestaiagentslatestaiagents
5

production-rag-checklist

Provide a production deployment checklist for RAG systems covering pipelines, monitoring, and security.

Official
Advanced
TheGreatLTheGreatL

prompt-caching

Implement multi-level caching strategies for LLM interactions.

Community
Advanced
KappTech88KappTech88

long-context

Extend transformer context windows to 32k+ tokens using RoPE, YaRN, ALiBi, or Position Interpolation.

Community
Advanced