Building a Second Brain — Voice → Notion → AI Workflow
How I turned Soundcore Work recordings into a self-organising Notion system — diary, ideas, and to-dos auto-sorted by Claude, summarised monthly. A no-code pattern that removes the friction of thinking out loud.
The Soundcore Work voice recorder sends transcripts straight into Notion — powerful on paper, messy in practice. After two weeks of use, my workspace was littered with hundreds of raw recording pages. Diary entries, product ideas, and to-dos all tangled together, none of them searchable, none of them reviewed.
The fix: treat the recorder as an input stream, not a destination. I wired Soundcore → Make.com → Claude → Notion so every new recording is classified, routed to the right database (Diary / Ideas / To-dos), and cleaned from the raw inbox — with a monthly summary generated automatically on top.
The critical design decision was idempotency. Version one had no dedup layer, so every time the scenario ran it re-processed the same pages, duplicating entries. The rewrite introduced a Processed Log database: before handling any record, the workflow checks it, filters out anything already seen, and logs new items at the tail of every branch. Without that, automation becomes a hazard.
Stack: Soundcore Work (¥24,990), Notion (free tier), Make.com (free tier), Claude API (a few hundred yen per month at my volume). No servers, no scheduler of my own — the whole thing runs on free infrastructure and one AI call per recording.
The unlock was not the tech; it was removing the friction of thinking out loud. I now talk to the recorder whenever something crosses my mind, knowing Notion will sort it. The monthly summary has quietly become the most valuable artefact — a mirror of what I actually cared about that month, not what I thought I did.
Full step-by-step implementation (Notion DB schema, Make.com scenario, Claude prompt, and the seven things that tripped me up) is on note.com — link below.