The honest starting point
AI is good at compression, structure and first drafts. It is not yet good at judgement, accountability or the things your business will be sued over. Knowing the difference is most of the design problem.
Where Twikkie uses AI today
- **CV summarisation** — a structured digest of every CV so the recruiter compares like with like.
- **Interview question generation** — drafted from the JD, then edited by the hiring manager.
- **Document summarisation** — a TL;DR on long policy or contract documents.
- **Internal Q&A** — the Twikkie Assistant answers product questions on twikkie.com.
- **Smart drafts** for the CMS, contract reminders, and case-management triage.
Where we deliberately don't use AI
- We don't auto-reject candidates.
- We don't auto-approve absences or expenses.
- We don't auto-classify performance.
- We don't generate audit log entries from inference.
The pattern: AI prepares the work; a human decides the work; the system records the decision.
Three principles we hold ourselves to
1. Show your work. Every AI suggestion has a source and is editable.
2. Refuse off-topic. Our chatbot will redirect rather than hallucinate.
3. Keep the human in the loop on outcomes. AI never writes the audit log on its own.
If those constraints sound limiting, they are — on purpose. The goal isn't impressive AI. It's a quiet, reliable platform that earns the boring kind of trust.