Every AI feature in rinkats has the same shape: the model reads, drafts, and stops. The profile draft waits for a person to accept it; the summary, the score and the questions are reads for a person to use. Nothing a candidate could see and nothing that moves a pipeline is written by a model alone.
What it reads and drafts
- Resume parsing. An uploaded resume becomes a structured profile draft: headline, summary points, skills with the months of experience the resume supports, and the timeline of roles. You review it section by section and keep your own version wherever you prefer it.
- Candidate summary. A one-line read, strengths with the evidence cited, a career arc, watch-outs, and a stability read based on the length of each stint. Verification flags call out things worth a question, such as overlapping dates, and they are flags, never verdicts.
- Screening against a requirement. A fit score with the reasoning shown. The score is deliberately loud in the interface because it is the one thing that should pull your eye. It never moves a card.
- Interview questions. Questions grounded in the candidate's own history and the requirement, with a place to record the answers and, when you ask for it, an evaluation of the recorded responses.
What it never does
It does not reject anyone. It does not email a candidate. It does not advance or drop a submittal. It does not share what it learned about your candidates with any other organisation.
Your keys, your models
rinkats runs on the model keys you bring. A recruiter can carry their own key, the organisation can hold shared ones, and usage is visible per key so there are no surprises. Any AI feature can be switched off for the whole organisation, and the tool keeps working without it.