I Read a Lot. I Remember Almost Nothing. So I Built This.

This is a series about building MarrowMinded — one AI lesson a day from books I’ve already read, made with Claude Code. Notes on the product, the process, and the tools along the way.

The series so far

1. I Read a Lot. I Remember Almost Nothing. So I Built This. — the origin story (you’re here)
2. Which AI Writes the Best Lesson? I Made Them Grade Each Other — four models grade each other’s lessons
3. The Email My App Never Sent — a daily email digest, from a one-inbox MVP to the full n8n flow

I finish a book. I feel good about it. Two weeks later, I can barely recall what it was about.

That’s the problem. And it’s not a lack of effort — I read consistently. The problem is the format. You read a book once, in big chunks, and then… nothing. The ideas don’t stick.

I’d read about the Ebbinghaus forgetting curve before — the idea that we forget roughly 70% of what we learn within 24 hours. I knew about it. I just didn’t do anything about it.

So I built MarrowMinded. The name comes from the idea of extracting marrow — the most nutritious part of the bone. Books have marrow too. Most of what surrounds it is just filler.

What It Does

MarrowMinded gives me one lesson per day from a book I’ve already chosen. Not a summary. Not a chapter. One concept, structured into seven sections:

• A short quote from the text
• The key thesis in one sentence
• The idea explained in depth
• A specific example from the book
• How it applies to my life today
• A concrete task for the day (5–30 minutes)
• A reflection question

That’s it. One lesson. Every day.

How does it look?

Why One Lesson?

Business books are mostly padding. An author has 40 pages of insight and writes 300 pages of anecdotes around them. The anecdotes make it readable. They also make it forgettable.

The insight from Effortless or Naval’s Almanack is there. The signal is real. But the format — read it once, all at once — is broken.

One focused concept a day, repeated over months, beats a 3-hour reading binge I won’t repeat or remember. Drip beats flood.

How I Built It

The stack is deliberately minimal. One Python file (FastAPI + SQLite), one HTML file with vanilla JavaScript. No framework, no build step. Books are plain .txt files. Lessons are generated once by AI and cached as JSON — so on day one of any book, there’s no API cost at all.

Multi-provider AI support under the hood (Claude, GPT, Gemini, Groq), but Claude is the default. The prebuilt lessons cost me around $50 in tokens total. The whole thing took 10–12 hours to build, mostly evenings.

It runs on Render’s free tier.

The Part That Surprised Me

I use it every morning. That’s the metric I actually care about.

Building something you genuinely use changes everything about how you build it. You cut the features you don’t need because you’d notice if they were wrong. You keep it simple because complexity would slow you down before breakfast.

I have eight books loaded in right now — a mix of English and Polish titles. Every morning, one lesson, one task. That’s the whole thing.

---

Do you have a reading habit? And more importantly — do you actually remember what you read?

If you’re curious how it works, have questions about the stack, or want to build something similar — ask. And if you’d like to take a look at the repo, just say so.