process-transcriptlisted

# Process Transcript Two halves. **Part A** is the file-based vault pipeline that runs on `.txt` files synced into `/Users/alexhedtke/My Drive/Plaud/` -- this is what the scheduled watcher triggers. **Part B** is the Plaud cloud API over MCP, for reaching recordings that have not landed as files or that Alex wants to query directly. # Part A: Vault pipeline ## Steps ### 0. Canonical name mapping Plaud transcripts frequently mis-transcribe names. Before processing any transcript, apply these corrections throughout the text: Before processing any transcript, check the People/ folder in the Obsidian vault for canonical spellings. Common Plaud mis-transcriptions include phonetically similar substitutions (e.g., "Linda" for a name ending in "-inda", "Bryce" for "[Friend]"). To build the correction table: 1. Glob `/Users/alexhedtke/Exobrain/Areas/Relationships & Community/People/*.md` 2. Use those filenames as the canonical names 3. Apply phonetic matching when Plaud produces a name that's close but not exact Also normalize variations of the same person to one canonical name for People/ notes (e.g., don't create both a nickname file and a full name file). When in doubt, use the fullest version of the name that exists in the People/ folder. Always check for existing People/ notes with similar names before creating a new one. ### 1. Find unprocessed transcripts - List all `.txt` files in `/Users/alexhedtke/My Drive/Plaud/` - Read `/Users/alexhedtke/Documents/Exobrain harness/