Retrieval-Augmented AI Engineering

Description

Language models do not know your company. Retrieval-augmented generation (RAG) fixes that by finding the right passages in your own documents and asking the model to answer from them, with sources. It is the most common pattern in real AI products, and also the one that is most often built badly. This course teaches you to build it properly and to measure whether it works.

The pipeline, piece by piece

• Embeddings and vector search: what an embedding is, similarity and nearest neighbor search, vector stores in practice and metadata filtering.
• Preparing documents: loading and cleaning, chunking strategies, chunk metadata and citations, and keeping an index up to date.
• Better retrieval: keyword search with BM25, hybrid search and rank fusion, query rewriting and reranking, and measuring retrieval quality with real metrics.
• Grounded answers: prompting with context and citations, conflicts, gaps and refusals, conversations and follow-up questions, and evaluating answer quality.

Real retrieval models, offline generation

The retrieval lessons use real open models from the sentence-transformers library (an embedding model and a cross-encoder reranker) that run on your machine and are downloaded once by the library, so the search results you see are the real thing. For the answer step the course provides an offline stand-in model, so you need no API key and no account, and the printed outputs are reproducible.

How the lessons work

Each lesson gives a realistic problem, a precise explanation with a diagram and the limits of the method, a code demo that you run, exercises with hints and a short summary. You are encouraged to compare methods on the same questions, which is why the course includes labeled test questions.

The project

The final module builds a Company Knowledge Chatbot over a knowledge base of 32 documents (company documents in plain files) with 40 labeled test questions: you ingest and index the documents, retrieve and answer with citations, handle follow-up questions, and evaluate and improve the system. You get a template repository, a worked solution repository, the knowledge base and the question set, plus a smaller lab corpus for the earlier lessons.

What is included

Lessons on embeddings, document preparation, better retrieval and grounded answers, a knowledge base with labeled test questions, a smaller lab corpus, a guided company knowledge chatbot project with template and solution repositories, and a quiz for every module.

Who it is for

Every company wants an assistant that knows its own documents, and retrieval-augmented generation is how it is built. You will not only build it here, you will learn to measure it, which is what most tutorials skip and what separates a chatbot people trust from one they abandon. If you want to build the AI assistant that businesses actually ask for, enroll now.