How Felix works
Last updated August 1, 2026
Felix is the coach inside Elovox. Felix is not a person; it's AI. Your feedback and scores are generated automatically by machine-learning models, not written by a human listening in. That's worth understanding before you lean on any of it.
What actually happens to a practice run#
- Speech-to-text turns your recording into a transcript with timings.
- A large language model reads that transcript plus the measured metrics (pace, fillers, pauses) and writes the coaching, the scores, and the summary.
- If you practiced with the camera on, a small number of still frames are also read by the model so it can comment on posture and expression.
The specific providers are named in the Privacy Policy, which also explains what is and isn't kept.
It can be wrong#
Both the transcription and the coaching make mistakes. The transcript can mishear you. The feedback can be confidently inaccurate, inconsistent between two attempts at the same thing, or just unhelpful. The scores are a practice aid to measure yourself against over time, not an objective assessment of your ability, and not a prediction of how a real interview, pitch, or audience will go.
Not professional advice#
Elovox is a practice tool, not a professional service. Nothing Felix says is professional, medical, legal, financial, or career advice. Don't make a decision that matters on the strength of it. Use it as one input, alongside people whose judgment you trust.
On the camera feedback#
When you turn the camera on, the model comments on visible delivery: posture, gestures, eye contact, expression. It is reading readily visible behavior to coach you, and it can misread you, especially if you move, present, or express yourself differently from whatever it treats as the norm. It does not identify you, and it builds no face template. See the voice & camera notice. The camera is off unless you turn it on.
On Felix's spoken take#
The short take at the top of a report is written by the same model from the finished analysis, not from the recording, and it is held to about sixty words. That is a summary of a summary: it says what matters most, and it can leave out something the full report covers. The written report underneath is the complete version, and the take is always there as text, so nothing depends on the audio. His voice is generated speech; see the privacy policy for who processes it.
Questions#
If anything about how Felix works is unclear, email support@elovox.app.