Build an assistant that answers from your own documents and shows where every answer comes from. You turn Larkfield's policy documents into a searchable knowledge base, measure what retrieval finds and misses, and make the assistant say "the documents do not say" when they do not.
Read every lesson as a guest, or sign in to enroll and track your progress.
Enroll to save the lessons that you complete. It is free.
By the end of this course, you will be able to:
A language model knows a lot, but it does not know your company’s documents, and it does not know which version is in force today. Ask it about Larkfield’s return policy, and it usually says that it does not know. Sometimes it answers with confidence, and it is wrong. It never shows where an answer comes from.
Retrieval-augmented generation fixes part of this. The application first finds the passages that matter, then asks the model to answer from them and to say where each claim comes from. Done well, every answer can be checked. Done badly, the assistant cites the wrong version, misses the table that held the answer, or follows an instruction hidden in a document.
In this course, you build Larkfield’s policy assistant step by step. You start with plain keyword search and a set of real questions, including questions the documents cannot answer. Then you add careful chunking, embeddings, hybrid search and reranking, and you measure each change instead of guessing. Finally, you keep the knowledge base honest as documents change, are deleted or are only for staff.
You do not need an account or a paid key. Search runs on your own computer, and every answer step has recorded responses from real models. Diagrams and explorers carry the main ideas, so the text stays short.
You completed SQL and Data Preparation for AI and Building Reliable Applications with LLM APIs, or you have the same skills. You do not need an API key or a paid account. Search runs on your computer with SQLite and small open models, every answer step has recorded responses, and live calls with a free local model or your own key are optional.
Courses to take first: SQL and Data Preparation for AI, Building Reliable Applications with LLM APIs.
People who can prepare data with SQL and Python and have built an application on a model API, as in Building Reliable Applications with LLM APIs, and now want the model to answer from their own documents. Developers, data scientists and professionals moving into AI engineering.
You can study the ideas of this course without writing code. Read modules 1 and 4–6 (about 10–13 hours), answer their knowledge checks, and get a free concepts certificate. You can come back for the practice later.
The paths for people who lead or use AI take this course this way. See the paths
Experienced AI practitioners designed this course. AI helped us draft the content; people chose the teaching approach and the examples, and reviewed every lesson. The outputs that you see are the actual results of running the code. How we make our courses
Understand what retrieval adds to a language model, and what it cannot fix. Define the documents and the questions, including the ones the documents cannot answer, and build a keyword-search baseline before anything else.
Turn documents into chunks you can trust. Extract text from Markdown, HTML and PDF, split it with care for headings and tables, and give every chunk an ID, a version and a source location.
Compare ways to find the right passage. See what embeddings capture, test keyword, vector and hybrid search on exact terms, paraphrases and rare codes, and measure what reranking and query changes add.
Write answers that stay close to the evidence. Choose passages within a budget, link every claim to a passage and check it in code, and abstain when the documents are missing, conflicting or out of date.
Measure each part of the pipeline separately. Use recall at k for retrieval and separate checks for correctness and citation support, then locate each failure and change one thing at a time.
Keep the knowledge base correct and safe over time. Update and delete documents so that old text disappears from answers, carry access rules through every step, defend against instructions hidden in documents, and hand the system over.
Three end-to-end case studies on open data. Answer questions from public guidance with citations, compare search methods on technical documentation, and answer from regulations that change over time. Complete at least one for the certificate; each one is a portfolio piece.
Complete at least one case study for the certificate. The others are optional. What each one covers
Use every module in one practice project. You collect evidence, and check it against the course outcomes.
Realistic projects from start to finish, on real data, with every output explained. Each one makes a portfolio piece. Complete at least one for the certificate; the others are optional.
Answer questions from public guidance4 hr
Build a cited question-answering assistant over open government guidance pages, with abstention for questions the guidance does not cover and a measured evaluation.
Search technical documentation2 hr
Compare keyword, vector and hybrid retrieval with reranking on openly licensed technical documentation, where exact names and paraphrases both matter.
Answer from versioned regulations3 hr 30 min
Answer questions about rules that change over time, retrieve the version in force on a given date, and cite it.
Loading the reviews…
Took this course? Sign in to review it. Learners can write a review after they complete 25% of the lessons.
