personalized autocomplete

an autocomplete that knows you really well — not BCI (though that would be cool) but from context about you generally or things you were just doing. specifically considered for meetings: a Granola-type deal where after a meeting you write up notes and it autocompletes as soon as you write enough because it was listening.

issues identified: might not be good or might take a lot of time (would basically need thought autocompletion), and it only saves very little time per completion. depends on the me model for deep personalization and could integrate with the always-on AI assistant for meeting context.

spreadsheet evaluation: originality 6/10, excitement 6/10, MVP 4-8 weeks. competitive landscape: Granola does meeting transcription + AI notes, GitHub Copilot autocompletes code, Gmail Smart Compose predicts email text, Lex.page does AI writing completion — none combine deep personal context (writing history, meeting context, personal knowledge graph) into autocomplete. quality bar is very high — bad autocomplete is worse than none. could prototype as a fine-tuned model on writing corpus + meeting transcripts. tech depth 4/10, labeled [ARCHIVE] ("interesting but small").

timeline

• [2026-02-06] captured — meeting notes use case, issues identified
• [2026-04-10] google sheets evaluation — 6/10 originality, 6/10 excitement, [ARCHIVE] tier