The Concept: Context without Search

The inception of PULSE came from a deep frustration with traditional note-taking apps. In a world of folders, tags, and complex graph views, the actual act of thinking often gets buried under the administrative burden of organizing. I wanted something that felt like a chat with my own shadow—a single, infinite vertical stream where the UI doesn’t ask you where to put a note, but instead tells you what you’ve thought about this before.

AI Brainstorming: From Static to Fluid

The brainstorming phase was heavily assisted by AI to explore the boundaries of Semantic Retrieval. Instead of “searching” for notes, we explored the idea of “beating” notes—hence the name Pulse. Each entry is a heartbeat.

The core challenge discussed with AI models was the latency of vector embeddings. We debated:

• Local vs Cloud: Should we use transformers.js to run embeddings in the browser or hit an API?
• Distraction vs Utility: How many matches are too many? AI helped simulate scenarios where real-time suggestions could become noise rather than a signal.

Implementation: The Vector Heartbeat

Building PULSE required a tight integration between the editor and the database. Using Convex, I implemented a system where every few keystrokes (debounced), the partial text is vectorized and compared against thousands of previous “heartbeats”.

• Editor: A custom TipTap implementation optimized for a “chat style” flow.
• Backend: Convex Vector Search allowed for sub-second matching, making the context suggestions feel like they were appearing just as the thought was forming.

The Reality Check: Feeling the Noise

Test phase was brutal. Once I had “touched” the app and started using it for real daily logs, I hit the first wall: The Over-Prompting Trap.

Initially, the app would show matching notes for every single sentence. It felt like someone whispering in my ear while I was trying to talk. I realized that semantic relevance doesn’t equal cognitive relevance.

The Pivot: I had to implement “Soft Context” rules—prioritizing recency and setting high similarity thresholds. I also realized that the “Infinite Flow” is great for capture, but terrible for review without an AI-driven summary layer.

Verdict: The idea is good, but the UX needs to be “polite.” It needs to be there when you need it, and invisible when you don’t. It’s a tool for non-linear thinkers, but it needs linear constraints to be useful.