Filler-Free

Filler-Free: Live is a real-time AI speech coach built on Agora's Conversational AI Engine. You explain something out loud — a bug you fixed, a project, an interview answer — and the AI interrupts you immediately, in under 8 words, the moment you say "um," repeat yourself, or ramble. Then it goes quiet the instant you're being clear. Silence is the reward for a clean explanation.

The goal: self-monitoring your own speech while you're talking is hard, and existing tools — recordings, playback — give feedback too late to actually build the habit. This closes that gap by correcting you mid-sentence, while the habit is still forming, instead of after the fact.

How it's built: instead of building a custom speech pipeline, I configured Agora's Conversational AI Engine to handle the barge-in — turn-detection, noise suppression, streaming transcription, low-latency TTS all come from Agora, configured rather than reimplemented. The actual product is a system prompt that tells the agent exactly when and how tersely to interrupt.

What makes it more than solo coaching: the coach can call real tools mid-conversation to check its own live stats and queue follow-up practice. A "Go Live" mode turns solo practice into a coached group session — a mentor can co-host, an audience can watch, and a second AI coach joins the room with a group-aware prompt so it corrects the speaker's filler words live for everyone to hear, not just in a private one-on-one channel. A headless CLI companion exercises the same backend for demos and testing.