生成式AI与RAG工程 Certification
构建基于检索的、面向生产的高质量大型语言模型应用
Key facts
- Level: Professional
- Field: Artificial Intelligence & LLMs
- Estimated study time: about 30 hours
- Credential price: $213
- Exam: 66 questions · 90 minutes · pass mark 75%
About the 生成式AI与RAG工程 certification
掌握基于大型语言模型的生产级生成式AI系统背后的工程学科。本专业认证从如何在推理阶段引导模型开始：提示技术（零样本、少样本、思维链）、系统提示与角色设计，以及控制确定性与多样性平衡的解码参数——温度、top-k、top-p。 课程的核心是端到端的检索增强生成（RAG）：加载与切分文档（切块大小、重叠与语义策略）、将文本嵌入以用于检索、使用正确索引（HNSW、IVF）将向量存入向量数据库，以及用余弦相似度和点积度量相似性。你将深入学习检索本身——稠密向量、稀疏/BM25词汇检索、混合融合，以及使用交叉编码器重排——然后组装一个遵守Token预算、避免「迷失在中间」失效模式的上下文窗口。 最后三分之一是将演示与产品区分开来的关键：通过引用为答案提供依据、诊断并缓解幻觉、评估RAG流水线的忠实性与答案相关性（RAGAS）、接入函数/工具调用与结构化JSON输出、添加护栏与安全过滤器，以及通过缓存控制成本。学完后，你将能够设计、构建、评估并调试一个有依据的RAG应用，并为每一个架构选择进行辩护。
What you will learn
The official 生成式AI与RAG工程 study course covers:
- Prompting & Decoding Control — Steer an LLM at inference time: zero/few-shot and chain-of-thought prompting, system prompts, and the decoding dials — temperature, top-k, top-p.
- The RAG Pipeline End to End — The full architecture: offline ingest (load, chunk, embed, index) and online query (embed, retrieve, rerank, generate), plus chunking strategy.
- Vector Databases & Retrieval — Storing and searching embeddings: HNSW and IVF indexes, similarity metrics, and dense vs sparse vs hybrid retrieval with reranking.
- Grounding, Hallucination & Evaluation — Context-window assembly, hallucination causes and mitigation, citations, and measuring RAG quality with faithfulness, relevance, and RAGAS.
- Tools, Structured Output, Guardrails & Cost — Function/tool calling, structured JSON output, guardrails and safety against prompt injection, and controlling latency and cost with caching.
- 动手实验——构建有依据的 RAG — 一个指导性顶点项目：在 Google Colab 中端到端构建一个可工作的、有依据的 RAG——摄取和分块，使用 FAISS（平面 + HNSW）嵌入和索引，检索并生成带引用的答案，然后用 recall@k 评估并添加护栏。免费运行。
Prerequisites
Frequently asked questions
- Is the 生成式AI与RAG工程 certificate verifiable?
- Yes. Every issued Hootix Academy certificate carries a unique credential code that anyone can verify online.
- How is the 生成式AI与RAG工程 exam structured?
- It is a 90-minute ID-verified multiple-choice exam of 66 questions; you need 75% to pass.
- Do I need to buy the course to take the exam?
- Each certification is sold three ways: the exam on its own, the online course on its own, or both together as a Course + exam bundle (the bundle price is the course price plus the exam price).
- How long does the 生成式AI与RAG工程 course take?
- About 30 hours of self-paced study.