The call is the visible part. The rating pipeline that runs after it is what turns 300 conversations into a shortlist you can act on before lunch. Here is exactly what happens in the sixty seconds after a Veytrix call ends.
Step 1 — The transcript is finalized
During the call, speech is transcribed live (Sarvam AI handles Indian languages and telephone audio). At hang-up, the full speaker-separated transcript — who said what, in order — is stored with the call record.
Step 2 — Summary and structured fields
A language model reads the transcript and produces a short summary plus the structured fields you asked for: notice period, expected CTC, current location, willingness to relocate — whatever your use case defines. Free-form conversation in, clean columns out.
Step 3 — Scoring against YOUR criteria
This is the part generic transcription tools don’t do. You define the rating template — say communication, relevant experience, notice-period fit, intent — and the LLM scores each finished call against those dimensions, with a short justification for each score. Ratings are consistent across all 300 calls because the same rubric grades every one; no interviewer mood, no 6pm fatigue.
Step 4 — The ranked view
The Ratings page sorts candidates by overall score. Click any row → full transcript, summary, per-skill scores. Export the lot to Google Sheets for the hiring manager. Recruiter time goes where it matters: the top of the list.
Can you trust LLM scores?
Trust them the way you trust a good first-round screener: excellent at consistent triage, not the final decision. Spot-check transcripts for the first campaigns, tighten your rubric wording, and the correlation with human judgement gets strong — while the cost per screen stays around ₹15. That trade is why screening is the first thing teams automate.
See a rating generated on a real call in the next ten minutes: create a workspace, run a test call, open Ratings. Or read the screening walkthrough first.