RAG
Retrieval-Augmented Generation (RAG) is an AI technical framework that combines information retrieval with the generative capabilities of large language models. Its core mechanism involves retrieving relevant documents or passages from an external knowledge base before generating a response, and providing them as context to the model. This allows the model to draw on current or domain-specific knowledge, effectively mitigating hallucination issues and knowledge cutoff limitations inherent in large language models.
Core Facts
The core idea of RAG is to retrieve relevant information from an external knowledge base as context to inject into the prompt before the large model generates a response.
The core of GraphRAG is integrating knowledge graphs into the traditional RAG pipeline, organizing information in the form of triples (two entities and their relationship).
Prompt注入（Prompt Injection）是针对大语言模型应用的安全威胁，攻击者通过在输入中嵌入恶意指令试图覆盖系统原有行为约束
Timeline (last 90 days)
RAG's effectiveness depends heavily on retrieval quality: if the reasoning behind a decision exists only in Slack conversations and is not indexed, or if PR descriptions are vague, retrieval cannot recall the key passages.
If the knowledge base itself is chaotic, outdated, or mutually contradictory, the retrieved context will mislead the model, and output quality will instead decline.
Deckhand adopts a RAG (retrieval-augmented generation) architecture: it first vectorizes the user's question, retrieves the most semantically relevant passages from the knowledge base, and then hands them to the large model to generate an answer.
For scenarios like the IT help desk, where answers are standardized but accuracy requirements are high, controllable knowledge sources determine the final result more than model scale does.
RAG is still essentially question-and-answer; it does not proactively break down tasks, does not execute in multiple steps, and does not retain long-term memory across turns, which is the key dividing line between it and an Agent.
The Agent technical architecture is divided into three layers from bottom to top: the traditional question-and-answer mode, RAG (Retrieval-Augmented Generation), and Agent (including toolset, Memory, and Planning).
The mainstream implementations of RAG and job requirements are basically built around the Python ecosystem; although there are corresponding libraries on the TypeScript side, they still lag relatively in maturity and community examples.
AI/ML-related positions are still predominantly Python-based, and job descriptions in areas such as data science, model fine-tuning, RAG pipelines, and Agent framework R&D almost all require Python skills by default.
RAG pipelines usually involve steps such as document chunking, vector embedding, similarity retrieval, and prompt assembly.
In a RAG pipeline, context carries factual information, so a low hallucination rate is an advantage for a model; in open-domain question answering scenarios without retrieval support, a low hallucination rate becomes a shortcoming (higher refusal rate, lower coverage).
40 more timeline events
All Facts (20)
The core idea of RAG is to retrieve relevant information from an external knowledge base as context to inject into the prompt before the large model generates a response.
90%VerifiedThe core of GraphRAG is integrating knowledge graphs into the traditional RAG pipeline, organizing information in the form of triples (two entities and their relationship).
90%VerifiedPrompt注入（Prompt Injection）是针对大语言模型应用的安全威胁，攻击者通过在输入中嵌入恶意指令试图覆盖系统原有行为约束
90%VerifiedLlamaIndex (formerly GPT Index) focuses on data indexing and Retrieval-Augmented Generation (RAG) scenarios.
90%VerifiedRAG（检索增强生成）架构由 Meta AI 研究团队在 2020 年提出
90%VerifiedCursor's AI agent capability uses RAG (Retrieval-Augmented Generation) technology to inject relevant code context into large language models to achieve precise task execution.
90%VerifiedThe typical workflow of RAG (Retrieval-Augmented Generation) consists of four steps: document chunking, vector embedding, similarity search, and context injection.
90%Verified混合检索（Hybrid Search）将向量检索与BM25关键词检索的结果通过RRF（倒数排名融合）等算法合并，兼顾语义理解与精确匹配
90%VerifiedPerplexity AI's underlying architecture uses RAG (Retrieval-Augmented Generation) technology, crawling web content in real time and combining it with LLM output to produce structured answers with cited sources.
90%VerifiedRAG（检索增强生成）技术架构将文档切片、向量化、存入向量数据库，查询时先检索相关片段再交给模型生成回答
90%VerifiedRetrieval-Augmented Generation (RAG) anchors facts by incorporating real-time web search with cited sources, but its effectiveness is influenced by factors such as search result quality, document parsing accuracy, and the model's comprehension capability.
85%VerifiedThe answer quality of RAG depends largely on the retrieval step rather than on the model's capabilities themselves.
85%VerifiedThe core logic of RAG is to slice documents and store them as vectors, then retrieve relevant fragments through semantic similarity at query time and inject them into the context.
85%VerifiedAI evaluation targets are divided into three layers—large language model evaluation, knowledge base (RAG) evaluation, and agent (Agent) evaluation—forming a progressive relationship of increasing complexity and knowledge depth.
80%VerifiedRAG (Retrieval-Augmented Generation) solutions based on vector databases are better suited for handling document-type knowledge, but often fall short in retrieval precision and real-time performance for highly structured engineering facts.
80%VerifiedThe working principle of RAG is divided into a retrieval phase (converting the question into a vector to recall relevant content from the knowledge base) and a generation phase (inputting the recalled context together with the question into the model to produce an answer).
80%VerifiedThe RAG workflow is divided into offline and online phases: offline, documents are chunked and converted into vectors stored in a vector database; online, questions are converted into vectors and the most similar text segments are retrieved via ANN algorithms and concatenated into the context.
80%VerifiedThe essence of RAG is decoupling parametric knowledge (knowledge solidified in model weights) from non-parametric knowledge (an enterprise's own externally updatable knowledge base).
80%VerifiedRetrieval-Augmented Generation (RAG) splits codebases into semantic segments via vector databases and dynamically retrieves the most relevant content to inject into the context when an AI handles specific tasks.
80%VerifiedVector memory databases convert historical conversation content into high-dimensional semantic vectors, and during new conversations dynamically retrieve relevant memory fragments via similarity search and inject them into the context window.
80%