When you answer a question in Claryve, a lot happens before you see feedback.
This post walks through the pipeline — from audio capture to the structured coaching note that appears on your screen.
Step 1: Transcription
Your answer is transcribed in real time using a streaming speech-to-text model. We preserve timing metadata so the feedback layer can reference specific moments in your answer.
Transcription runs client-side for low latency. The raw transcript is never stored — only the structured feedback derived from it.
Step 2: Structured Analysis
The transcript is passed to a language model with a prompt that extracts:
- Clarity — Is the answer easy to follow?
- Specificity — Does it include concrete examples or metrics?
- Structure — Does it follow a recognizable pattern (STAR, problem/solution, etc.)?
- Completeness — Did you address all parts of the question?
Step 3: Feedback Rendering
The structured output is rendered as a coaching note with inline references to your transcript. You can see exactly which sentence triggered each piece of feedback.
AI feedback is a signal, not a verdict. Use it to identify patterns across sessions — a single session's feedback is less meaningful than a trend over five sessions.
What We Don't Do
We don't score you on a 1–10 scale. Scores feel precise but they're not — they collapse nuance into a number that's easy to fixate on. Instead, we surface patterns and let you decide what to work on.