Course Knowledge Bases for Teachers and Trainers: Sources, Integrity and Selling Your Notes

A course knowledge base turns your syllabus, lecture notes, handouts and reading lists into an AI tutor that students can question directly, with every answer traced back to the specific slide, page or paragraph it came from. Done properly, it reduces repetitive email questions, gives students a revision tool that actually reflects what was taught, and avoids the two big fears teachers have about classroom AI: made-up answers and students outsourcing their thinking entirely.

Why teachers and trainers are building course knowledge bases

Every lecturer and corporate trainer has answered the same question by email a dozen times:

Every lecturer and corporate trainer has answered the same question by email a dozen times: where does this topic appear in the slides, what did we actually say about exceptions to the rule, is this in scope for the exam. A general-purpose chatbot cannot answer those questions reliably because it was not trained on your specific syllabus, and it will happily invent a plausible-sounding answer that has nothing to do with what you taught. A course knowledge base solves this differently: you upload your own material, the system indexes it for retrieval, and every answer the AI gives is grounded in passages it actually found in your documents, with the source shown alongside the answer.

This matters just as much for corporate and vocational trainers as for university lecturers. A compliance trainer running sessions on anti-money laundering rules, or an apprenticeship provider covering a technical syllabus, faces the same problem: learners want a quick way to check a detail after the session ends, without waiting for the next cohort or digging through a shared drive of PDFs.

What goes into a course knowledge base

The strongest course knowledge bases are built from the material you already have, not from a rewritten textbook. Useful source documents typically include:

• The syllabus or module handbook, including learning outcomes and assessment criteria
• Lecture slides and speaker notes, ideally exported as PDF or Word rather than screenshots
• Seminar handouts, worked examples and problem sheets
• Past exam papers with model answers or marking schemes (where you are comfortable sharing these)
• Reading lists with summaries of key chapters or articles
• Transcripts of recorded lectures, if you record sessions

Scanned handouts and photocopied worksheets are common in teaching, and extraction quality on these matters more than people expect. If your source material includes scanned pages or handwritten annotations, it is worth reading our guide to OCR and extraction quality before you upload a large batch, because a poor scan produces a poor chunk, and a poor chunk produces a poor answer regardless of how good the retrieval system is.

Keeping the AI tutor honest: grounding answers in your material

The single biggest risk with any AI tool in education is confident fabrication. A student asks a question, the model does not find a clear answer in the source material, and rather than saying so, it produces something that sounds right. The fix is structural rather than a matter of prompting harder: the system should retrieve the relevant passages first using hybrid search (combining keyword matching with meaning-based search, which our explainer on hybrid search covers in more depth), then generate an answer that cites exactly which passage it used, or decline to answer when nothing relevant is found.

If you want to sanity-check whether an AI tutor is actually behaving this way before you roll it out to a class, our guide on checking AI answer citations walks through the practical steps: click through to the cited passage, confirm it supports the claim, and watch for answers that cite a document but not a specific location within it.

Academic integrity: AI tutor versus homework machine

The distinction that matters for integrity is not whether students use AI, but what the AI is allowed to draw on and what it is asked to do. A course knowledge base that only contains your own syllabus material, and that is explicitly scoped to answer

The distinction that matters for integrity is not whether students use AI, but what the AI is allowed to draw on and what it is asked to do. A course knowledge base that only contains your own syllabus material, and that is explicitly scoped to answer what did we cover and where, behaves very differently from a general model asked to solve the coursework question itself. Some practical ways teachers narrow the scope:

• Exclude solution sets to live assessments from the knowledge base, even if you include past papers from previous years
• Set the assistant's purpose to explain concepts and point to sources, not to produce finished answers to current assignment questions
• Keep a private, cohort-only base for anything tied to live assessment, and a separate public or shared base for general subject material
• Tell students explicitly, in the module handbook, what the tool is for and what counts as acceptable use, in line with your institution's academic integrity policy
• Review a sample of logged questions periodically to check the tool is being used for revision rather than as a way to generate submitted work

This is also where citations double as an integrity feature rather than just an accuracy one: when every answer points to a specific slide or page, students get into the habit of checking the source rather than copying the generated text verbatim, which is a better outcome than a tool that gives a fluent paragraph with no way to verify it.

Private cohort bases versus public course libraries

Not every course knowledge base needs the same level of access control. A base built from this term's live coursework questions should almost always stay private, queried only by you and your students through an invite link or institutional login. A base built from general subject material, an introductory statistics course or a recurring professional qualification, can reasonably be made more widely available, and even generate income if you are the one who wrote the material.

Selling your course knowledge beyond the classroom

If you have spent years refining a set of course notes, a revision guide for a professional exam, or training material for a vocational qualification, that material has value to learners well outside your own classroom. A public knowledge base lets other students or trainees pay per question to query your notes directly, with citations back to your original material, rather than you selling a static PDF that goes out of date. On Kopik specifically, creators set their own price per question and keep 70% of it, while the platform handles extraction, indexing and the question-answering layer.

This is the same model covered in more detail in selling your expertise to AI agents and selling expertise online as a paid AI knowledge base, both of which apply directly to teachers and trainers: the mechanics of pricing, positioning and keeping a public base current are the same whether the author is a tax adviser, a solicitor or a lecturer. A sensible starting point is to publish a free sample base covering one topic, see whether learners actually use it, then expand and price the fuller version once you know the demand is real.

Setting up a course knowledge base in practice

The practical steps are the same whether you are building a private revision tool for forty students or a public resource for a professional qualification:

1. Gather your source documents (PDF, Word, Markdown or plain text all work) and remove anything tied to live, unmarked assessment
2. Create a base at /dashboard/bases/new, upload the documents, and let the system extract, chunk and index them for hybrid search
3. Decide whether the base is private (only you and your API keys or invited students) or public and listed in the catalogue
4. Test it yourself with the questions students actually ask, and check the citations point to the right slide or page before you share access
5. Choose how students will query it: through a simple web interface, embedded in your institution's portal via the REST API, or through an MCP server if you want it usable from tools like Claude or ChatGPT
6. Set a review cadence, once a term is usually enough, to re-upload updated slides and retire outdated past papers

If your course notes are long and structure-heavy, with numbered sections, definitions and worked examples, how the documents get split into chunks noticeably affects answer quality. Our guide to chunking strategies for RAG is worth a read before you upload a large handbook, particularly if you notice the assistant mixing up two different sections in one answer.

Data protection and safeguarding considerations

Course material itself is rarely a data protection problem, but it is easy to accidentally include personal data when you upload a folder in a hurry: marked scripts with student names, attendance registers embedded in a spreadsheet, or feedback comments that identify individuals. Before you upload anything in bulk, strip out personal data that is not needed for the knowledge base to function, and keep assessment-related material in a private base rather than a public one. Our UK GDPR checklist for RAG knowledge bases covers the specific questions worth asking, including who can see query logs and how long they are retained, which matters more when some of your users are under 18.

Getting the balance right

The teachers who get the most value from this approach treat the AI tutor as an extension of office hours, not a replacement for teaching. It answers the repetitive questions, points students back to the exact passage where a concept was introduced, and frees up real contact time for the things a retrieval system genuinely cannot do: judging whether a student has actually understood something, and having the conversation that follows when they have not.

None of this requires a large technical project. The underlying mechanism, document in, question answered with a citation, is the same whether it is deployed for a single module or sold as a professional revision resource. Starting small, with one module's worth of notes and a handful of test questions, is the fastest way to find out whether it genuinely saves you and your students time before you commit to anything larger.