How AI helps — and its limits
We use AI carefully, and only in ways that keep a person in control of every decision that matters.
It suggests, it never decides
After you review your report and confirm submission, Sarva automatically analyses the text you confirmed to suggest a preliminary incident category and to check whether the submission contains relevant incident information. This happens as a core function of the service. You are not asked for a separate AI-assessment permission. A trained human always reviews every suggestion before it is used. The labels are not legal decisions and do not determine whether a crime occurred. The service classification is anti-Hindu hate; another religion is never suggested as the report category.
It always shows its evidence
Every suggestion is shown next to the exact quote from your account that it was based on, so reviewers (and you) can see exactly why it made that suggestion.
It never judges you
AI in this service is never used to decide whether a crime occurred, to score how credible you are, or to decide what happens to your report next. Those are always human decisions.
Voice help is optional
If speaking is easier than typing, you can use voice input. Audio and transcript data are processed by a voice service to provide the conversation. Sarva does not store raw audio. What you say becomes an editable draft — not official evidence — and nothing is submitted until you review it, confirm submission, and choose sharing. Sarva does not infer emotion, credibility or risk from your voice.