I read a lot of non-fiction. And I have a problem: I can’t remember half of what I read.
Not because I don’t try. I highlight, I take notes, I make Anki cards. But the connection between reading and retention is broken. You read a great chapter, close the book, and a week later you’re struggling to explain the main argument to a friend. The ideas were clear when you read them — they just didn’t stick.
Existing tools don’t help. Summarizers give you bullet points that strip out the nuance. Flashcard apps treat every book the same regardless of complexity. And book clubs are great but require other people to be on the same schedule.
So I’m building something that bridges the gap.
What Ziline Does
The core idea is simple: the tool should adapt to how deep you want to go, not force you into a one-size-fits-all mode.
Five-layer architecture:
- Surface layer — what is this book about? Who’s it for? A 30-second scan.
- Structure layer — the argument map. How do the chapters connect? What’s the through-line?
- Concept layer — key terms, frameworks, mental models. With definitions that link to other books.
- Application layer — how do you actually use this? Exercises, prompts, “try this tomorrow” suggestions.
- Synthesis layer — how does this connect to what you already know? Cross-book links, contradictions, gaps in your knowledge graph.
Each layer is optional. Want to skim? Stay at layer 1. Want to really own the material? Drill down to layer 4. The AI tracks where you spend time and adapts — if you keep skipping the exercises, maybe the book’s not the right fit or you need a different entry point.
The Hard Parts
Adaptive depth model — figuring out when to push deeper and when to back off. Too aggressive and it feels like homework. Too passive and it becomes another summarizer nobody uses.
Cross-book knowledge graph — this is the long-term bet. If you’ve read 20 books on decision-making, Ziline should surface the contradictions between Kahneman and Taleb, not treat them as separate silos.
Retention without gamification — I don’t want streaks or badges. The motivation should come from actually understanding things better, not from keeping a number alive.
Where It Is Now
The PRD has been through two rounds of challenger review and defense response. The design is getting sharper. Core architecture is settled, but the actual product doesn’t exist yet — no landing page, no prototype, just a lot of thinking on paper.
I’m intentionally going slow. This is the kind of product where getting the interaction model wrong means nobody uses it, no matter how good the AI is. I’d rather spend two months on the PRD than six months rebuilding after launch.
ziline.cc — placeholder for now. Real thing coming eventually.