Memory & learning

[ComputelessComputer]

Keyword boosting

When users correct mistranscriptions (proper nouns, jargon, company names), the system should remember and boost those words in future transcriptions. Most STT providers already support keyword boosting or custom vocabularies — we need to expose it in a thoughtful way.

Implicit learning from edits

When a user corrects "Hyperno" to "Hyprnote" in a transcript or summary, that correction should feed back automatically. Track edits across surfaces (transcript editor, summary editor) and build a user-level dictionary without requiring explicit configuration.

Merged from

• implement keyword boosting #3512 — implement keyword boosting

Related

• Transcript editor #2547 — transcript editor (editing UX where corrections happen)

[ComputelessComputer]

Technical context from merged issues

From #3512 (keyword boosting):

• Most STT providers support keyword boosting / custom vocabularies natively
• Three approaches considered: implicit learning from edits, learning from transcript/summary edits, global dictionary/vocabulary
• Open questions: scope (note-level vs user-level vs workspace-level), how explicit in UI vs fully automatic, how to surface and manage learned keywords
• Previously considered tools: supermemory, momo