Introduction
For years, we've been collecting information from everywhere-GitHub repositories, YouTube videos, articles, meeting notes, and random thoughts. The problem isn't capturing information anymore. It's finding the right piece of knowledge when you actually need it.
During the The Hangover Part AI Hackathon, I wanted to explore a different question:
What if AI didn't just store memories, but actually understood how they're connected?
That idea became Memzee, an AI-powered second brain that converts scattered information into an interactive knowledge graph. But building the product taught me much more than implementing features.
The Problem
Traditional note-taking applications are excellent at storing information. They're much less effective at answering questions like:
- "Where did I learn this?"
- "How are these two ideas connected?"
- "Which project used this technology?"
- "What did that YouTube video teach me?"
Most systems still rely on folders, tags, or keyword searches. Human memory doesn't.
The Idea Behind Memzee
Instead of organizing information manually, Memzee focuses on understanding relationships. Every memory becomes part of a connected graph.
Users can:
- Capture thoughts
- Import GitHub repositories
- Import YouTube videos
- Explore connected entities
- Recall information using natural language
The graph continuously evolves as new information is added.
Why Cognee?
Cognee handled the part that interested me the most—treating information as structured knowledge rather than isolated text. Instead of building a retrieval pipeline from scratch, I could focus on creating the experience around it:
- Interactive visualization
- Timeline navigation
- Knowledge exploration
- Recall interface
- Import workflows
That allowed me to spend more time improving usability.
Biggest Engineering Challenges
1. Knowledge Graphs Are Harder Than They Look
Generating nodes is easy. Keeping the graph readable isn't. I quickly realized that as more information entered the system:
- Nodes overlapped
- Edges became cluttered
- Navigation became confusing
That forced me to rethink layout algorithms and graph interactions instead of simply rendering more data.
2. Deleting Knowledge Isn't Simple
Removing a memory wasn't just deleting one record. The graph also had to stay consistent. That meant rebuilding relationships without leaving orphaned nodes or broken edges. Sometimes "forgetting" required more engineering than "remembering."
3. Imports Needed Context
A GitHub repository isn't useful because it's code. It's useful because of the knowledge inside its README. Similarly, YouTube videos became valuable once transcripts could be converted into connected information.
The challenge wasn't importing content. It was making imported knowledge discoverable.
What I Learned
This project changed how I think about AI products. I learned that:
- AI isn't only about generating text.
- Retrieval matters as much as generation.
- Visualization dramatically improves understanding.
- Small UX improvements often create more value than adding another AI feature.
- Building an MVP means choosing what not to build.
What I'd Build Next
If I continue developing Memzee, I'd focus on:
- User authentication
- Private knowledge graphs
- PDF imports
- Website imports
- Better graph clustering
- Semantic filtering
- Collaboration
Final Thoughts
The biggest takeaway wasn't learning another framework. It was realizing that the future of personal knowledge isn't about taking more notes. It's about helping people connect what they already know.
That idea is what inspired Memzee.
Try It Out
You can explore Memzee and view the live project here: Memzee
Top comments (1)
The second-brain idea gets interesting when capture and retrieval have different standards. Capturing everything feels productive, but retrieval needs trust, source, and a way to ignore stale notes.