Turning Psychology Book Notes into a Second Brain with an LLM Wiki
Knowledge only truly enters your brain when books begin to converse with each other.
The other day, I was chatting with a friend about "boundaries." I had read more than a dozen books on the topic, and yet I couldn't put together a single coherent take. Too many books, too many interpretations all came rushing in at once, with no way to lay them on top of each other. That moment made me realize: yes, I had finished the books, but the knowledge hadn’t actually entered my brain.
Over the past five years I've read close to a hundred books, and the genre I devoured most eagerly was psychology — boundaries, attachment theory, trauma, self-compassion… After each one, I felt like I had gained something new. To deepen the impression, I even built a digital bookshelf in Notion to record reading notes, highlights, and summaries (you can see this article). I also tried using GoodNotes on my iPad to pull every book's take on the same concept onto a single page, hoping to assemble the full picture. But that approach was incredibly time- and effort-intensive, and I still couldn't truly understand any abstract concept thoroughly. Books that didn't actually "land" after I finished them were pretty discouraging.
How an LLM Wiki Got Books Talking to Each Other
Through my work, I started getting more involved with AI — and gradually came to appreciate what LLMs (Large Language Models, the core technology behind ChatGPT, Claude, and their kind) are genuinely good at: opening dozens of books at once, laying them flat side by side, finding the vocabulary they share, and mapping the connections between them. For a human, that’s an enormously draining cognitive task. For an LLM, it’s almost effortless.
Around that time, AI researcher Andrej Karpathy shared the idea of an “LLM Wiki” — feeding a large collection of personal materials into an AI so it can help you build a private knowledge base you can explore through conversation. The moment I read it, I knew this was exactly what I’d been looking for.
So I built one with Claude. I called it my “Psychology LLM Wiki,” and I’ve been adding books to it one by one — feeding in my reading notes, personal reflections, and the sentences that moved me most — around 20 books so far. But it’s more than a smarter book-summary tool. It’s closer to a living knowledge network, a place where books can genuinely speak to each other.
🧩 Benefit 1: Concept Summaries — Abstract Ideas Finally Have Dimension
The Wiki organizes things along two axes: one indexed by book, the other indexed by concept — like "boundaries," "codependence," "self-compassion"… Each concept automatically gathers every book that mentions it, recording how different authors interpret it.
Take "boundaries." After ingesting more than a dozen relevant books, the Wiki distilled a shared core that cuts across cultures and contexts: "Setting a boundary is about restraining yourself, not controlling the other person." At the same time, it laid out each book's distinct angle.
These views had previously been scattered across separate pages, each in isolation. It would have taken a huge amount of effort, plus a strong memory, for me to overlay them in my head. But in the LLM Wiki, all those voices show up on the same page. I can finally see "boundaries" — not as a single book's definition, but as a three-dimensional shape collectively traced out by a whole group of books.
💬 Benefit 2: Cross-Book Q&A — The Wiki Pulls Answers from the Whole Shelf
When something is gnawing at me, I can drop the question straight into the Wiki and have it pull together an answer from every book I've read.
A while ago, I asked it: "Why do I feel guilty?" After flipping through several books at once, it gave me an answer roughly like:
- Levin in Setting Boundaries: "Guilt is a sign that a boundary is taking effect."
- Wu Zhihong in Emotional Blackmail: "Guilt is often an emotion someone else hands to you."
- Neff in Self-Compassion: "Let's stay gently with the feeling of guilt."
Each author has their own stance and tone. When several voices arrive simultaneously from different angles, the emotion immediately gains depth. It feels less like reading a summary and more like sitting in the middle of a circle of authors, listening to them converse — or debate.
🧠 Benefit 3: Synthesis — The Wiki Grows New Knowledge of Its Own
This is the most fascinating, and most unexpected, ability of the whole Wiki: it can synthesize.
Get Yu-Ching Lin’s stories in your inbox
Join Medium for free to get updates from this writer.
Each time I add a new book, I ask it: "Does this conflict with or echo what's already there? Is there a new synthesis?" It then connects the dots on its own, and new perspectives sprout up.
For example, different psychology books use slightly different terms to describe similar phenomena — "gaslighting," "mystification," "double-bind." The concepts overlap heavily, but the meanings differ slightly. The Synthesis sorted out the source and meaning of each, and simplified them into a relationship like "double-bind < mystification ≈ gaslighting." That insight grew out of where the books intersect; no single book on its own could have produced it.
What's more, once the concepts in the Wiki accumulate to a certain density, it starts analyzing me back. One day, across the seven or eight concepts I'd been reading about, it spotted the same pattern: a sensitive child, in an environment that wasn’t safe enough, inventing tools to survive out of whatever materials were on hand — just given different names by different schools of thought. For the first time, I felt like I could really see the motivation behind all the years I'd spent reading.
That's what makes the LLM Wiki so compelling for me: it isn't a fancier summarizer. It’s a knowledge network that grows, ruminates, and reads me into itself.
If You Want to Try It
If you'd like to try this yourself — I don't have a technical background, and building it really wasn't as complicated as I'd imagined. AI luminary Andrej Karpathy recently shared a recipe: with one prompt and Claude Code, you can spin up your own LLM Wiki. The steps are simple — just follow along:
- Create a project folder — for example, one called
Bookshelf. - Create a raw materials folder — add a
rawsubfolder inside it. - Import your materials — drop your reference materials into
raw, or import them via MCP. Word, Excel, PDF, Markdown, and TXT are all supported. - Grab the core prompt — copy this prompt template and paste it into Claude Code.
- Add your customization — append the following to the prompt:
- “You are my dedicated LLM Wiki Agent. Please fully implement the above as my second brain, and walk me through it step by step. Start by creating CLAUDE.md as the Schema file, containing the complete operating rules; then create index.md and log.md to define the folder’s naming conventions, and produce a first sample summary for me to confirm. From now on, every interaction must follow the Schema rules, and the entire conversation should be in Traditional Chinese.”
From there, you’ll see Claude Code start automatically organizing the relationships between pieces of knowledge and creating cross-referenced documents. If you want to visualize the knowledge graph, just open the project folder in Obsidian and everything becomes clear at a glance.
Knowledge Is a Net That Keeps Being Rewoven
The way this psychology LLM Wiki works is actually a lot like the brain itself. Every book we read, every pit we fall into, every culture we experience adds a new node, links it back to the old nodes, and slowly weaves the unique person we are today.
The LLM Wiki helps with the parts I don’t have enough cognitive bandwidth to do alone — connecting nodes from different books and letting the conversation actually happen. Having this Wiki saves me effort, generates more distinctive nodes, and lets me weave my knowledge net denser and wider.
I hope you’ll use LLM + Wiki to build your own second brain too.
Post-Promotion Reflections
To reach a wider audience, I shared the article on Facebook. The response was heartwarming — some said they wanted to try it themselves or recommend it to friends, others felt it was something everyone should have, and a few had already experimented with it in different fields. Through this process, I also made a few new observations:
- Imagining Applications Across More Fields. This approach holds potential well beyond psychology. Fields like academia, government policy, legal documents, business strategy, and even yoga therapy — where years of course notes on anatomy and kinesiology can be systematically cross-referenced — could all benefit from this method. It helps organize perspectives from multiple sources and spark new lines of thinking.
- The choice of input material matters. Using raw book content directly as input data can raise copyright concerns and run into token limitations. That’s why I focus on entering my reading reflections, notes, and the passages that personally resonated with me — rather than feeding the AI the original text. This way, the AI can understand what I truly value, and may even be able to analyze me through the lens of what I’ve chosen to highlight.
