Building the knowledge base
The knowledge base is the material the teaching draws from. Ground every later phase in it rather than in your own training. Two paths, chosen by whether the learner already has content in mind.
Contents
- Path A: find content to study
- Path B: use content the learner already has
- Fitting material into context
- Tooling for hard-to-reach content
Path A: find content to study
Search for material relevant to the learning goal — recent blog posts, talks and videos, research papers, books. Favor substance over marketing; skip vendor content that sells rather than teaches.
Aim for breadth: around ten candidates, each with a title, a one-line summary, and a link. Then narrow with the learner. If the first sweep is too narrow or off-target, widen the theme and search again rather than forcing the original framing.
When the list gets unwieldy, group it by type (article, paper, video, book) so the learner can see the shape of what's there.
Path B: use content the learner already has
The learner may arrive with specific books, bookmarks, or files. Take them as the spine of the knowledge base, then integrate any supplementary sources around them rather than letting one source dominate.
Fitting material into context
Before committing to a large set of sources, check it realistically fits what you can hold. When it doesn't, two moves:
- Prioritize and drop the weakest sources.
- Split the topic into sub-topics and learn them in separate passes.
A smaller, fully-read knowledge base beats a large one you can only skim.
Tooling for hard-to-reach content
Some material can't be read directly and needs extracting first. Verify access to every source; for what's locked away, these help:
- Video transcripts — transcript extractors (e.g. TubeTranscript) turn talks into text.
- Code repositories — repomix packs a repo into a single LLM-friendly file.
- Websites — webmix packs a site into one file.
- Books — for DRM-free or owned books, OCR/export tools (e.g. Kindle-AI-Export) produce readable PDFs.
These illustrate the kinds of tools available; the specific one matters less than getting clean text you can actually teach from. Drop or flag anything you genuinely can't access — never teach from a source you only imagine you read.