#RAG
287 related articles
Bilibili's 748-Episode AI Large Model Zero-to-Hero Course: A Full Breakdown from RAG to Agent in Practice
Full breakdown of Bilibili's 748-episode AI large model course: from fundamentals to prompt engineering, then RAG, LangChain, and Agent development with three enterprise projects.
XERJ: A New Way to Search Data for AI, Reducing Token Consumption by 40x
XERJ is an open-source AI data search tool on GitHub that uses autoindex to automatically index code, docs, logs, and PDFs, cutting token consumption by 40x vs grep. Compatible with Elasticsearch, supporting RAG and agent memory scenarios.
Don't Rush to a Vector Database: First, Nail Down Your Retrieval Baseline
Building a RAG assistant doesn't require starting with a vector database. Learn how to establish a measurable retrieval baseline using existing full-text search and SQL filtering, identify vocabulary gap scenarios through an evaluation set of fifty real questions, and then decide whether to introduce vector retrieval and hybrid search.
RAG in Production: Should You Really Use Frameworks Like LangChain?
A developer shares production-grade RAG engine experience: when to use LangChain/LlamaIndex frameworks and when to write native code. Document parsing, retrieval fusion, and observability each have their best choices.
What Is RAG? Retrieval-Augmented Generation Explained Through an n8n Workflow
Using a library analogy to explain RAG (Retrieval Augmented Generation) and demonstrate document loading, embedding, Pinecone storage, and retrieval generation in an n8n workflow.
LLM Wiki vs RAG: Real-World Test Shows Replacing Retrieval with a Wiki Is Actually Slower
Vellum Labs benchmarked an LLM wiki vs RAG/agentic retrieval on the 1,176-line itsdangerous codebase: the wiki was 55% slower and 34% pricier per first-round question, but answered the fifth question nobody had written down.
After the RAG Pipeline Works: Permission, Traceability, and Data Consistency Challenges in Enterprise Retrieval-Augmented Generation
Getting a RAG pipeline running is only the beginning. This article analyzes the three hurdles of enterprise retrieval-augmented generation: document permission control, answer traceability and auditing, and multi-source data consistency—and why these can't be solved by prompts or pipelines.
Chunking Is the Product, the Model Is Just the Engine
RAG performance depends on chunking strategy, not the embedding model. This article examines structure-aware chunking, table and footnote traps, adversarial evaluation, and why your chunker is the real product.
Malav Naik: From Data Engineering to Knowledge Graph-Driven AI Systems
Malav Naik, from Gujarat, India to Dublin, with five years of data engineering at S&P Global, now builds AI automation at Trinity Care Nursing Homes; his SkillBot combines RAG with a 2.3M-triple RDF knowledge graph, achieving 97% factual accuracy.
Why I Abandoned Answer Generation After Building a RAG Search Engine for 10,000 Documents
A developer shares lessons from building a RAG search engine for 10,000 Russian religious documents: structure-aware chunking, hybrid retrieval, RRF fusion, Cohere reranking, and why he ultimately abandoned LLM answer generation for pure semantic search returning original sources.
RAG Evaluation and Hallucination Guardrails: How to Detect and Block AI Hallucinations
Learn how to evaluate RAG systems with context relevance and faithfulness, use LLM judges to block hallucinations, and build regression test sets to prevent degradation.
Extracting Data from Long PDFs: Why Direct LLM Reading Fails and the Agentic IDP Approach
Why feeding long PDFs whole to an LLM fails: four failure causes, chunking's blind spots, and the Agentic IDP approach—page-by-page parsing, cross-page extraction, Schema definition, cost comparison, and data integrity checks.
Making Web Pages Survive RAG Chunking: Writing Strategies for AI Citations
RAG systems evaluate content by passage, not page. Learn to simulate chunking, detect weak passages, and win AI citations with self-contained answers, query-style headings, and JSON-LD.
When AI Confidently Gives Wrong Answers, the Problem Isn't Always the LLM
AI confidently gives wrong answers — the problem may not be the LLM itself. A developer breaks down three failure layers in a brand marketing copy system: brand context, retrieval, and generation, arguing that blindly tweaking prompts only treats symptoms while locating the failing layer is the key to reliable AI applications.
Structuring Data for AI Agents: Five Approaches and a Selection Guide
There's no single best way to structure data for AI Agents. This article compares semantic models, knowledge graphs, RAG, Markdown+grep, and open formats, drawing on research from dbt, Microsoft, and Chroma to help you pick the right data architecture based on question type.
Copying Is Not Grounding: ASC Exposes the Faithfulness Reward Loophole in Long-Form RAG
New research reveals RAG faithfulness rewards can be gamed by copying. The Authored-Sentence Credit (ASC) method moves rewards from answer-level to sentence-level, cutting error rates and boosting supported sentences across four models and five corpora.
How Many Tokens Does It Cost to Feed a Webpage to an LLM? 20 Real Pages Tested
20 real pages tested: median raw HTML is ~128K tokens; LLM-oriented Markdown saves 98% vs HTML and 63% vs full-page text. Main content extraction drives 51% of savings.
A Complete Guide to RAG Evaluation Metrics: Retrieval, Generation, and Operations
A systematic breakdown of RAG evaluation metrics: retrieval (Hit rate, Recall@k, MRR, nDCG), generation (faithfulness, answer relevance), and operations (cost, latency, storage), with examples and tooling.
EduRAG Intelligent Q&A System: A Hands-On Breakdown of a Heima AI Agent Project
A breakdown of the Heima EduRAG Intelligent Q&A System course, covering LangChain, Milvus vector database, hybrid search, and reranking to solve factual hallucinations in the IT domain, with job market insights and a learning path.
RAG vs Fine-Tuning: Which Should Your Enterprise Chatbot Use?
RAG or fine-tuning—which one should you choose? This article compares how both approaches work, their use cases, and cost structures in plain language, plus a five-question decision checklist to help enterprises pick the right technical path for their chatbot.