Overview
Brain Spill started as a personal frustration. I think better out loud than I type. But voice memos are a graveyard — you record them, never listen back, and forget everything.
The insight was simple: transcription isn't enough. You need structure. Brain Spill records your voice, transcribes it, then uses AI to extract action items, key ideas, and themes — formatted as clean notes you actually want to read.
The Problem
Voice memos have a 90% abandonment rate among knowledge workers I surveyed. The friction isn't recording — it's retrieval. A wall of unstructured transcript is just as useless as no notes.
What I Built
Core Loop
- Press record — speak freely, no structure needed
- Stop recording — transcription begins immediately
- AI structures the output — headings, action items, key points extracted
- Clean notes land in your library
Technical Architecture
Transcription: Whisper API for high-accuracy speech-to-text. Handles Polish, English, and mixed-language input.
Structuring: GPT-4 with a carefully tuned prompt chain. The prompt instructs it to preserve the speaker's voice while imposing structure — not rewrite the content.
React Native app:
- Expo Audio for recording with waveform visualization
- Background processing with push notification on completion
- Offline note storage with sync on reconnect
- Tag and search functionality
Design Philosophy
The interface needed to feel frictionless to the point of disappearing. There's one button. That's intentional.
The hard design decision was the waiting state — AI processing takes 10–30 seconds. I chose a minimal animated waveform over a progress bar. It felt more honest: you're watching something think, not watching a fake loading bar.
What I Learned
AI latency is a UX problem before it's a technical one. Users will tolerate 30 seconds if the interface communicates that something meaningful is happening. They won't tolerate 10 seconds if it feels broken.
Prompt engineering is real engineering. The difference between a well-structured note and a confused summary is a dozen iterations on the system prompt.
Status
Live on Google Play. iOS submission in progress.