From Paper Piles to Knowledge Graphs: A Bookshelf Epiphany

There is this co-incident that happened last week (and now I get to talk about it :) ) I was organizing my documents, papers, books - I have a personal collection of about 450 books - and at the same time, I had a podcast playing in the background. It was about knowledge graphs. I know it sounds super geeky, but it ain’t, stay with me.

So there I was, surrounded by piles of books and loose pages, trying to make sense of my own tiny library. I generally group my books by genre, and then sub group some by author, but then I realized that a few were just lying around because I didn’t know where to place them, that’s when something clicked.

The podcast host was talking about how AI struggles to find connections across big piles of information, as in how it sometimes gives half-right answers or just makes things up. And suddenly, I realized my books were doing the same thing to me! I knew I had read something about a particular topic, but I couldn’t remember which book, or what chapter. Everything was technically “there”, but nothing was actually connected.

And then it hit me, what if I had a map of my bookshelf? Not just one that shows where each book is, but one that connects the ideas inside them. Like, this sci-fi novel imagines pilots navigating mental interfaces to fly spacecraft, and that aviation book explains how real-world cockpit systems evolved.

That’s essentially what a knowledge graph does, isn't it?

Instead of treating the content like a stack of pages to flip through, it treats it like a network of information. It’s like one of those detective boards in crime shows, with strings, thumbtacks, photos but instead of me manually connecting the dots, it’s all digital, and way smarter. 

Most people (myself included, until recently) assume that if you just plug an AI into a big collection of documents, it’ll magically know how to answer questions. That’s kind of true, especially with something called RAG (Retrieval-Augmented Generation), where AI searches for text snippets and tries to piece together an answer.

But in reality it’s not that seamless, ‘coz fundamentally the content is messy and  tangled,  like product manuals, internal wikis, support docs… or yeah, even my own bookshelf, simple keyword search doesn’t really work. The AI either throws wild guesses or, it apparently, sounds super confident while being totally wrong. 

That’s where the knowledge graph comes in, it brings structure. It lets the AI understand not just what things are, but how they relate.

And the best part is, we don't have to create a big, complex database to organize information. That used to be the way to do it. Now, we can make a “knowledge map” right inside the same search system that the AI already uses. There's no need for special new tools or extra work to keep it up to date.

Here’s how I’d like to think about it:

• First, AI reads your content and builds a kind of mental model, it breaks down each piece into smaller, meaningful chunks and links them based on who or what they’re about.

• Then, when someone asks a question, the AI picks out the important “things” in the question, which is the people, products, events and follows the links between them.

• Finally, instead of returning random pages, it tells a story backed by multiple, connected sources, almost like how I’d explain something to a friend using three different books from my shelf.

It’s like upgrading from a search bar to a thinking assistant.

That random afternoon of organizing papers while listening to a podcast taught me something unexpected, which is, if you want better answers, whether in AI or in life, sometimes all we need is a better-organized way of keeping things, be it books or data.