Skip to main content
← WorkClarifi Voice · AI Output Evals
Case study · AI · alignment

I engineered a linting framework that strips AI markers and enforces brand personas—evals for prose, not vibes.

Portfolio home, about, and case study copy rewritten under hire-facing rules—same pattern enterprises use for LLM-as-judge and alignment pipelines.

5 voice profiles · contextual lint · hire vs product tone

5 profiles
hire · product · brand hierarchy
6 lint rules
vague · sleaze · AI markers · density
2,021
User messages mined for voice gold

Lint model (v0)

No fixed word bans—contextual rejection when copy is too salesy, vague, flowery, or AI-marked for the surface.

  • Vague stack — 3+ low-specificity adjectives without mechanism
  • AI marker clusters — delve, landscape, leverage chains
  • Filler density — repeated ensure/robust/comprehensive in short spans
  • Sleaze / gimmick — manipulative edu-marketing tone

Profile hierarchy

Cross-rules govern all surfaces. Hire copy uses I voice on portfolio and resume. Product brands use we with empathy-forward Weable tone and enterprise-trust Clarifi split.

How I use it

Every portfolio page, case study, and chatbot response passes the same lint—fact-backed, I voice, no jargon meant for my repo. The CV filter doc is the gate before ship.