I Built an AI-Powered Learning OS Because I Was Tired of Forgetting Everything I Read

Most people read non-fiction the same way: skim chapters, highlight some quotes, feel productive for a day, and forget 90% within a week.

I've been there. I'd finish a 300-page book on strategy or psychology, feel like I learned something profound, and then two weeks later struggle to recall a single framework from it. The notes I took gathered dust. The highlights were buried in an app I never reopened.

Turns out, this isn't a discipline problem. It's a biology problem — and there's 140 years of research proving it.

So I built ReadGraph — a self-hosted AI learning system that turns non-fiction books into something you actually retain. Not another reading app. Not another AI chatbot. A complete learning OS that understands what you've read, finds what you're missing, tests your understanding, and maps how everything connects.

Eight major versions and thousands of lines of code later, here's the story — and the science behind it.

The Problem: Your Brain Is Designed to Forget

In 1885, German psychologist Hermann Ebbinghaus ran a series of experiments on himself — memorizing nonsense syllables and testing his retention at intervals. What he discovered became one of psychology's most replicated findings: the forgetting curve.

The data is brutal:

• Within 20 minutes, you forget roughly 42% of newly learned material
• Within 1 hour, you lose over 50%
• Within 24 hours, only about 33% remains — two-thirds is gone
• Within 1 week, up to 90% has faded

A 2015 replication study by Murre & Dros, published in PLOS ONE, confirmed Ebbinghaus was right — their modern results showed 31.7% retention at 24 hours, almost identical to the 1885 data.

This isn't about intelligence or effort. This is how human memory works by default. The forgetting curve is exponential and merciless — unless you do something to fight it.

That "something" is well-documented in learning science. The problem is that most reading tools don't implement any of it.

ReadGraph does.

The Science That ReadGraph Is Built On

Every major feature in ReadGraph maps to a research-backed learning technique. This isn't marketing — these are the same methods used in medical schools, language acquisition programs, and cognitive psychology labs.

1. Spaced Repetition — Fighting the Forgetting Curve

The most powerful weapon against forgetting is spaced repetition: reviewing material at strategically increasing intervals, timed to hit right before you'd naturally forget.

The research is overwhelming:

• A 2025 meta-analysis of 13 studies across 21,415 learners found a significant effect favoring spaced repetition over standard study techniques, with a standardized mean difference of 0.78 (p < 0.0001).
• Medical students using spaced repetition scored 58.03% vs 43.20% for control groups in a large-scale family medicine study (p < .001, Cohen's d = 0.62).
• Nursing students using spaced repetition flashcards were nearly 3x more likely to pass exams (OR = 2.84) and over 2x more likely to achieve the highest grade (OR = 2.31).
• A 2025 paediatrics study showed post-test scores jumping from 11.42 to 16.24 in the intervention group, with over 90% of students reporting improved retention.

How ReadGraph uses this: The vocabulary builder in the Communication Lab uses the SM-2 algorithm (the same one powering Anki and Duolingo) to schedule concept reviews at optimal intervals. Assessments are also timed to interrupt the forgetting curve at the right moment.

2. Retrieval Practice — The Testing Effect

You remember what you actively retrieve — not what you passively re-read. This is one of the most robust findings in learning research.

A systematic review of 50 classroom experiments involving 5,374 students found that 57% of studies showed medium or large benefits from retrieval practice. It worked across different education levels, content areas, and experimental designs.

The key finding: retrieval practice with feedback produces an effect size of g = 0.50 — a substantial improvement. Without feedback, the benefits diminish significantly. The feedback piece is critical.

How ReadGraph uses this: When you finish a chapter, ReadGraph generates proper assessments — multiple choice, short answer, scenario analysis — built from your actual reading. It grades honestly with a 75% pass threshold, identifies gaps, and gives detailed feedback. This isn't a quiz for fun; it's retrieval practice with feedback, exactly as the research prescribes.

3. Concept Mapping — Knowledge That Connects

Isolated facts are fragile. Connected knowledge is resilient. Cognitive psychologist Dan Willingham put it simply: "Understanding is remembering in disguise."

Research published in Instructional Science (2025) showed that retrieval-based concept mapping performed as well as traditional retrieval practice, with both significantly outperforming simple re-studying on transfer tests one week later. The concepts stick because they're anchored to other concepts.

A flipped classroom study comparing knowledge graphs to concept maps found that students using knowledge graphs showed better performance in both breadth and depth of subject knowledge. The semantic relationships between entities support deeper cognitive engagement and collaborative knowledge construction.

How ReadGraph uses this: Every concept, resource, book, and chapter becomes a node in an interactive 3D knowledge graph. The connections form automatically as you read. By your fifth book, you have a living map of how ideas relate across subjects — the kind of cross-domain insight that makes knowledge compound instead of sit in silos.

4. Self-Explanation — Learning by Articulating

A meta-analysis of 64 research reports found that self-explanation prompts produce an effect size of g = 0.55 — asking learners to explain concepts in their own words generates inferences about causal connections and conceptual relationships that passive reading never triggers.

The researchers specifically recommended that future systems explore computer-generated self-explanation prompts rather than relying on instructor-scripted ones.

How ReadGraph uses this: The AI mentor doesn't just answer questions — it asks them. It prompts you to explain concepts, connect ideas across chapters, and articulate your understanding. The gap analysis feature identifies where your understanding is shallow and pushes you to go deeper. This is self-explanation driven by AI, exactly the direction the research points toward.

What ReadGraph Actually Does

The science is the foundation. Here's what it looks like in practice.

Think of ReadGraph as a personal study partner that never gets tired, never judges you, and has perfect memory of everything you've ever studied.

Organize Your Reading

You add a book — type the table of contents or snap a photo of it (OCR handles the rest). ReadGraph breaks it into chapters and tracks your progress. Simple, but it gives you an honest view of where you actually are.

Automatic Knowledge Enrichment

When you add a book, ReadGraph automatically searches the web for articles, blog posts, case studies, and resources related to each chapter. These get processed and stored in your local knowledge base.

By the time you sit down to read Chapter 3, supplementary material is already waiting — alternative explanations, real-world examples, and context the author didn't include.

An AI Mentor That Actually Knows Your Content

This isn't ChatGPT with a "pretend you read this book" prompt. The AI mentor has real access to your notes, the enriched knowledge base, a knowledge graph of concepts, and your reading history. When you ask "How does this relate to what I read in Chapter 2?", it looks up what you actually studied and gives a grounded response with citations.

The mentor also renders visual explanations inline — flowcharts, concept maps, comparison charts, and diagrams — right inside the chat.

Gap Analysis

ReadGraph compares what you've actually read against what's in the knowledge base and identifies specific concepts you haven't encountered yet. It's like having someone who read the same book plus 50 related articles pointing out your blind spots.

Real Assessments with Honest Grading

AI-generated assessments — multiple choice, short answer, scenario analysis — grounded in what you actually studied. A 75% pass threshold. No sugarcoating. This is the retrieval practice piece that interrupts the forgetting curve.

A Living 3D Knowledge Graph

Every concept becomes a node. Every relationship becomes an edge. You can literally see how ideas cluster across books and subjects. The "everything is connected" moment is real — and it's backed by the concept mapping research above.

Study Planner

AI generates a realistic day-by-day reading plan with calendar and timeline views. Not generic advice — a plan that accounts for chapter length, complexity, and your reading pace.

Communication Lab

Practice applying what you learned through role-play scenarios. Pitch to a simulated investor. Negotiate a deal. Includes a vocabulary builder with SM-2 spaced repetition so terms actually stick in long-term memory.

Reflective Journal

A private journal with mood tracking and an AI companion. Learning hard material is emotionally taxing. The journal gives you a space to process that.

Dashboard That Tells the Truth

Reading streaks, time tracking, coverage charts, understanding metrics, and AI-generated insights. The honest mirror that shows whether you're progressing or just going through the motions.

The Benefits: Why This Approach Works

You Actually Retain What You Read

Spaced repetition + active recall + concept mapping + self-explanation. These aren't buzzwords — they're the four most evidence-backed learning techniques in cognitive science, and ReadGraph implements all of them in one integrated system.

Your Knowledge Compounds

Most reading is isolated — each book sits in a mental silo. ReadGraph's knowledge graph breaks those silos. By your fifth book, you have four books' worth of concepts already mapped, and the system shows you connections you'd never notice on your own.

You Get Honest About Your Gaps

Gap analysis removes the illusion of understanding. You learn where your real blind spots are, not where you think they are.

Your Data Stays Yours

ReadGraph runs entirely on your machine. Your notes, reading history, knowledge graph — all stored in a local PostgreSQL database. No cloud. No subscription. No risk of a startup shutting down and deleting your data. (Remember when Pocket shut down in July 2025 and 17 million users lost everything? That can't happen here.)

It Works for Any Non-Fiction

Psychology, history, science, philosophy, business, self-improvement — if it has chapters and concepts, ReadGraph can help you learn it deeply.

You Control Your AI Spending

Built-in budget controls let you set daily, weekly, or monthly spending limits. The system tracks every API call's cost and pauses when you hit your cap. Supports 300+ AI models including free ones. No surprise bills.

The Journey: From Weekend Project to Learning OS

ReadGraph didn't start as what it is today.

The early versions were bare-bones — a book tracker with a basic AI chat. Notes lived in browser storage. No enrichment, no knowledge graph, no assessments.

The middle versions added the pieces that made it actually useful: automated enrichment that finds supplementary material, a knowledge graph connecting concepts across books, and real assessments with honest grading.

The recent versions focused on depth and polish: the Communication Lab for practicing what you learn, spaced repetition for vocabulary retention, a journal for processing the emotional side of learning, rich visual responses (diagrams, charts, concept maps), a complete design system, budget controls, accessibility, and security hardening.

Eight major versions. Each built on the insight that reading and learning are different activities, and the gap between them is where the real opportunity lies.

Who Is This For?

• Startup founders trying to self-educate on business, finance, marketing, and leadership — fast
• Career changers reading their way into a new field
• Students who want to go deeper than their coursework
• Researchers building knowledge across papers and books
• Lifelong learners tired of reading books and forgetting them
• Anyone who reads non-fiction and wants to actually retain and apply what they read

If you've ever finished a book and thought "I feel like I should know more than I do" — that's the exact frustration ReadGraph was built to solve.

The Philosophy Behind the Design

Three principles guided every decision:

Local-First

Your learning data is deeply personal. It reveals what you don't know, where you struggle, and how you think. That data should live on your machine, not on someone else's server. ReadGraph is fully local — your database, your files, your control.

AI-Augmented, Not AI-Dependent

The system enhances your reading process — it doesn't replace it. You still read the book. You still take notes. You still think through problems. The AI finds supplementary material, identifies gaps, generates assessments, and visualizes connections. It's a force multiplier for effort you're already putting in.

Depth Over Breadth

ReadGraph is deliberately designed for one book at a time, read thoroughly. It doesn't optimize for reading speed or volume. It optimizes for understanding. One book read deeply with ReadGraph will teach you more than ten books skimmed without it.

What's Next

ReadGraph is a living project. Some areas being explored:

• Multi-book synthesis — generating insights across your entire library
• Obsidian Sync — exporting your knowledge graph directly to your Obsidian vault
• Deep Research — multi-agent research workflows
• Local LLM support — running models on your machine for complete privacy

The core thesis remains the same: the gap between reading and learning is solvable, and the right combination of AI, knowledge mapping, and active recall can close it.

Try It Yourself

ReadGraph is available now as a one-time purchase — full source code, no subscription, runs locally forever.

Get ReadGraph at readgraph.app

If you think this kind of tool should exist, I'd also appreciate your support on Product Hunt — it helps other learners discover it:

Vote for ReadGraph on Product Hunt

Final Thoughts

I built ReadGraph because I was frustrated with my own reading habits. I was consuming books but not learning from them. The highlights were piling up but the knowledge wasn't.

The research was clear: forgetting isn't a character flaw — it's biology. And the tools to fight it have existed in academic literature for decades. Spaced repetition. Retrieval practice. Concept mapping. Self-explanation. The problem was that no one had put them all together into a system designed for how real people actually read non-fiction.

What surprised me most was how much the knowledge graph changed my relationship with reading. Seeing concepts from different books cluster together — realizing that a chapter on negotiation tactics connects to a chapter on cognitive biases from a completely different book — that's when reading starts to feel like building something rather than consuming something.

If you've ever felt the gap between how much you read and how much you retain, you're not alone. That gap is exactly what ReadGraph was built to close.

One book, read deeply, is worth more than ten books skimmed.

ReadGraph is a self-hosted AI learning OS, currently at v8.0. One-time purchase. Full source code. No subscription. Runs on your machine forever.