Why RAG Matters
Retrieval-Augmented Generation (RAG) is the most practical technique for grounding LLM outputs in real data. The awesome-llm-apps repository contains 19 distinct RAG implementations, each solving a different challenge.
Basic RAG Chain
The foundational pattern: embed documents, store in a vector database, retrieve similar chunks at query time, and pass them as context to the LLM. Simple but effective for most use cases.
Corrective RAG (CRAG)
A self-correcting pipeline that grades retrieval quality before generation. If retrieved documents are irrelevant, CRAG triggers web search as a fallback. This dramatically reduces hallucination from bad context.
Autonomous & Agentic RAG
Autonomous RAG — The agent decides whether retrieval is needed, what to search for, and which strategy to use (vector, keyword, hybrid). Adapts behavior based on question complexity.
Agentic RAG Variants — Multiple implementations using different models (Gemini, DeepSeek, Cohere) and reasoning strategies.
Knowledge Graph & Hybrid Search RAG
Knowledge Graph RAG — Builds a knowledge graph from documents, enabling multi-hop reasoning and proper source citations. Every claim links back to a specific passage.
Hybrid Search RAG — Combines keyword search (BM25) with semantic search (embeddings) for better recall. Available in both cloud and local variants.
Vision RAG & Production Patterns
Vision RAG — Extends RAG to visual content — images, charts, diagrams. Uses multimodal embeddings to retrieve and reason over visual data.
RAG-as-a-Service — A production deployment pattern with API endpoints for document upload, collection management, and query.
RAG Failure Diagnostics — A diagnostic tool for identifying why RAG pipelines fail — irrelevant retrieval, chunking issues, embedding quality problems.
All 19 implementations are available as runnable Python applications in the awesome-llm-apps repository.