674 saved links, concentrated in 2024–2025 — the corpus captures RAG’s arc from hot new pattern to mature engineering discipline, and the 2026 saves show the conversation shifting toward context engineering as the umbrella term. Worth noting: this wiki itself is a retrieval system over your knowledge — the techniques on this page apply directly to the thing you’re building.
Related: AI Agents · LLMs · Azure · Data Science
Start here
- A Complete Guide to RAG in 2026 — the current-state overview.
- By-hand lecture slides: RAG, vector databases, agents — visual, beginner-friendly, downloadable.
- Mastering RAG: ~200-page guide to enterprise-grade systems.
- Weaviate’s free RAG book — written by their DevRel team.
- Jason Liu: building a terrible RAG system is easy — the anti-patterns; read this before building anything.
- @bclavie’s lecture — “best explanation of how to do RAG right” (your own note says so).
Fundamentals: embeddings & vector search
- LLM Embeddings Explained: visual and intuitive guide — you saved this twice, months apart; it’s that good.
- Embeddings are underrated — the case for embeddings beyond RAG.
- 6 embedding types and when to use them — sparse vs. dense vs. late-interaction.
- The Hitchhiker’s Guide to Vector Search and Vector Databases Explained in 3 Levels of Difficulty.
- Sinusoidal positional encoding and RoPE explainer — the layer below embeddings.
- Qdrant Academy — interactive vector-search courses.
Retrieval quality: the techniques that matter
The consistent message across your saves: nail retrieval first, generation follows.
- Jason Liu: RAG isn’t about better generation — nail retrieval and six low-hanging fruits RAG search leaves on the table.
- Chunking: Chunking Strategies for RAG · chunking mechanics: accuracy, context, speed · the “lost context problem” — why naive chunking breaks multi-hop questions.
- Reranking: Fowler/Subramaniam: the Reranker pattern · hybrid search + reranking with LanceDB · LLM-labeler to fine-tune a fast reranker.
- Advanced: hierarchical indices · 9 techniques to boost RAG · Hamel: late interaction models · RAFT: retrieval-augmented fine-tuning · embedding fine-tuning guide with LangSmith + RAGAS.
GraphRAG & knowledge graphs
Your second-densest RAG sub-topic — retrieval over structure instead of similarity.
- How GraphRAG Works Step-by-Step — start here.
- Graph RAG by Example and building a Graph RAG system step by step.
- Building knowledge graphs with LLM Graph Transformer · from unstructured text to interactive knowledge graphs.
- The quest for production-quality Graph RAG — the honest difficulty assessment.
- Guido van Rossum’s “structured RAG” package — LLM-extracted entities at ingestion instead of embeddings-only.
- PIKE-RAG (Microsoft): specialized knowledge & rationale augmentation.
- AI agents with Google Gen AI Toolbox + Neo4j.
Agentic RAG
Where your two biggest AI topics meet: retrieval as a tool the agent wields, iteratively.
- Agentic RAG: the survey — architectures and taxonomy.
- Building Agentic RAG Systems (video) and a full course on agentic RAG.
- AUTO-RAG: autonomous iterative retrieval — the LLM decides when to retrieve again.
- Beyond RAG: agent search with LangGraph.
- Agentic RAG + MCP servers: implementation guide.
- “Is context engineering the new RAG?” — the framing shift, from Microsoft’s Foundry IQ team.
Evaluation
- Hamel: modern IR evals for RAG — how to measure retrieval, not vibes.
- Auto-creating a RAG eval dataset from your documents.
- open-rag-eval — scalable RAG evaluation, LangChain connector.
- Hallucination detection techniques and evaluating modular RAG with reasoning models.
- RAG vs. Agent+RAG evaluation with TruLens.
RAG on Azure
The biggest platform cluster in your RAG saves — your Azure depth applied.
- Implementing agentic RAG on Azure and agentic RAG with Semantic Kernel + Azure AI Search.
- Build a RAG agent with Azure AI Agent Service SDK (Janakiram MSV).
- Azure AI Search + Docling for document ingestion.
- the official Azure AI Search RAG tutorial notebook.
- Cosmos DB series: movie chatbot with vector search + semantic cache · NoSQL + vector search + Azure OpenAI in Python · Cosmos embeddings generator.
- Simple RAG using SQL Server and OpenAI · automating RAG indexing with Logic Apps.
Case studies & patterns
- Uber Genie (your best real-world RAG story): RAG copilot saved Uber 13,000 engineering hours · Genie: the on-call copilot · how it was built — LangGraph, Qdrant, Gemini, Ragas.
- Fowler/Subramaniam: GenAI patterns series — RAG as an architecture pattern, including what lies beyond RAG.
- Context-Augmented Generation: enterprise refinement of RAG.
- Permissions (the enterprise gotcha): RAG without leaking sensitive data — OpenFGA + LangChain · users should only see results from documents they can access.
The “RAG is dead?” debate
- RAG isn’t dead — but vectorstore-as-default might be.
- Why RAG is still relevant — costs and hallucination control.
- “RAG is not really a solution” — the strongest counter-argument you saved.
- Jason Liu: predictions for the future of RAG.