The Problem with Traditional Note-Taking
I am an avid reader—at least, I would love to be identified as that. I love reading books ranging from spiritual topics and personal development to deep dives into technology. But along the way, I was faced with two major problems.
First, for books that explain heavy concepts, I found out I had a massive retention issue. I simply was not able to recollect the deep details inside those pages a few weeks later.
Second, I found it hard to grasp certain complex passages while reading. This made me resort to running Google searches mid-chapter, which constantly interrupted my reading flow.
To fix this, I decided to start taking notes of great points and storing them in my Obsidian vault. That did solve the first part to some extent, as I now had a repository to fall back on every now and then. But I still wasn't retaining the information in my actual memory. It felt like a glorified copy-paste of the book into my personal notes. Even worse, it did nothing to address my second problem of breaking down hard concepts in real-time.
This frustration led me to research better, more scientifically proven ways to learn and retain what you read.
What i Found
During my research process, I came across two powerful frameworks that perfectly articulated what I was experiencing: Schema Theory and the PACER Framework as explained by Dr. Justin Sung.
I came to understand that learning happens in two distinct phases: the consumption phase and the digestion phase. Looking back, my issue wasn't consumption; it was entirely about the digestion phase.
Dr. Sung states that you need different strategies to digest what you learn based on the type of material you are engaging with. The PACER system outlines exact routines for each information type: practice for procedural skills (Practicals), create and critique for comparing alternative approaches (Analogies), mind mapping for core relationships (Concepts), and storing raw data for Evidence. For Reference information—which consists of arbitrary facts like specific dates or precise code syntax—his strategy is to externalize it completely into a dedicated system like flashcards for spaced repetition so it doesn't clutter your working memory.
For me, my problem sat squarely within the A-C-E spectrum: Analogies, Concepts, and Evidence. I thought about manually forcing myself to follow through with these strategies, but I still faced heavily fragmented focus. I would read the e-books on my phone or tackle a hard copy, attempt to map out conceptual relationships using wordy text blocks in my Obsidian notes, and try to critique my understanding using various LLM tabs.
This setup caused a massive loss of focus. I would constantly deviate from reading to doing other random stuff online. Most importantly, I still didn’t have a clear mental picture of how things fit together because I am a highly visual learner—I love visual mappings and Excalidraw diagrams, for example.
So, I realized I had two options:
- Build a unified workspace that allows me to study, verify my understanding of analogies and concepts, add supporting evidence, create a beautiful visual map of relationships, and store them directly in markdown notes.
- Rely entirely on existing platforms like ChatGPT, Claude, or NotebookLM.
Ultimately, tools like NotebookLM didn't cut it for me. I don't like their rigid graph nodes; I prefer the flexible, visually pleasing layout that an open canvas like Excalidraw offers. Building a custom solution myself gave me the luxury of personalizing the environment and tweaking it to suit my exact cognitive needs.
The Architecture: Cognee Cloud and Excalidraw
Intending to build this solution myself, I immediately faced a classic AI development challenge: the context window bottleneck. Every single time I wanted to query an LLM to check my understanding of a concept, it had to load the entire chapter or book into its memory window. This consumed a massive amount of tokens, meaning more latency and higher expenses.
I needed a permanent, persistent memory setup for my application, and that is where Cognee fit into my pipeline.
Cognee is an open-source data management tool that structures text by extracting semantic triples. Instead of passing massive text files back and forth, it maps information into deterministic structures of [Source] ──[Relationship]──> [Target].
By pairing this graph processing framework on Cognee Cloud as my intelligent backend memory with an embedded Excalidraw Canvas on the frontend, I was able to design a dual-pane workspace. Users can load their e-books or text materials natively on the left pane, chunked down chapter-by-chapter to prevent cognitive overload, while sketching out their visual mental models on the infinite canvas to the right.
Verifying Understanding with Graph Layouts
Using a cloud-persistent graph completely solves my memory issue. It allows me to upload my source material exactly once. Cognee Cloud processes the book text, transforms it into an optimized knowledge graph, and hosts it securely.
This introduces a massive architectural win: it completely decouples the raw book file from the real-time validation loop. Once the graph is serialized into memory, the original heavy PDF or ePub file is irrelevant. The application never has to re-read or re-load the book ever again.
When I finish reading a chapter and draw my mental model on the canvas, I hit a button to verify my layout. The app converts my Excalidraw JSON elements into semantic triples and runs a lightweight graph traversal to mathematically compare my drawn map with the textbook's true graph layout.
If my understanding is flawed, the app calculates the variance and exposes my exact knowledge gaps visually:
- If I draw a relationship arrow backward, it flags red.
- If a critical connection is missing entirely, a dashed placeholder node appears to show the empty slot in my mental schema.
To guide me through these gaps, an integrated Socratic chatbot panel automatically slides open. Because it reads the exact missing links from the graph engine, it doesn't just blurt out the answer. It acts like a real tutor, asking brief, sharp questions to trigger cognitive accommodation: "Your concept layout for Account Addresses looks solid, but you are missing the cryptographic trigger between the Public Key and the Address. Look back at section 2—how do we get from one to the other?"
By preserving these verified chapter graphs in the cloud, the long-term vision for Mental Note goes even deeper: Cross-Domain Far Transfer. Because all my books live within the same unified graph ecosystem, the app will eventually be able to track network shapes across entirely different subjects. If I map out a brand-new concept that shares the exact same structural graph topology as a framework I studied weeks ago in a different field, the app can proactively surface that familiar analogy to my consciousness—allowing me to assimilate complex new ideas instantly.
Conclusion
My ultimate takeaway from this project is that note apps should not just be a storage vault. Instead, they should function like an active learning assistant and a Socratic teacher that actively helps you test your mental models.
The app is currently optimized as a zero-friction demo. There are no heavy database bottlenecks or sign-up requirements; your chapter notes and Excalidraw mental graphs are saved directly to local browser storage so you can test the validation loop instantly.
I would love to hear about other people's learning workflows! How do you currently break down complex concepts, and prevent yourself from falling into the passive copy-paste trap?
Top comments (0)