Voice profile

Out of the box, AI-generated CVs and cover letters sound generic — recruiters can spot them. A voice profile teaches every downstream agent (cv-adapter, cv-rewriter, qa-agent, linkedin-composer) what your real writing voice is, so the output sounds like you wrote it.

What “voice” means here

Four small documents the agents read:

  • Unified Writing Rules — vocabulary you use, words you avoid, sentence cadence, register (formal vs casual).
  • CV Writing Rules — bullet structure, action verbs you favor, ones you never use.
  • Cover Letter Writing Rules — opener pattern, paragraph density, signoff style.
  • Other Formats Rules — Slack-style notes, LinkedIn posts, etc. — applies wherever you write outside CV/CL.

The user-voice-creator agent generates these from your samples. You can re-edit any of the four documents directly in the UI.

Feeding samples

Settings → Voice Profile → Upload samples. The agent reads your samples (emails, blog posts, anything in your real voice) and synthesizes the four rule documents. Two samples minimum — one is too thin, the agent over-fits.

Each sample should be 200-2000 words of real prose. Don’t feed it AI-generated text from another tool — that contaminates the voice.

Auto-deletion of raw samples

The Privacy Policy commits to deleting your raw samples once the voice profile is built. After the agent succeeds, the route runs purgeVoiceSamples in a 1-retry loop. If the purge fails twice in a row, an admin gets paged via the data_deletion_incidents table — your samples remain visible in Settings until manually purged.

Re-building

Re-running the user-voice-creator agent (Settings → Voice Profile → Re-create) replaces the existing four documents. Add fresh samples first if your style has evolved.