echolag is a browser based coaching lab for real conversations. it runs immersive scenarios, tracks filler words and pace, and gives instant feedback so you can sharpen how you speak without the pressure of practicing on people.
we wanted a safe space to practice high pressure conversations. pitching, interviewing, or calming a rushed customer should not require asking friends to sit through awkward role play.
echolag runs immersive scenarios in the browser like the barista rush. it listens while you speak, highlights filler words live, and submits your take the moment the countdown hits zero so you stay in flow.
the frontend is next.js and react with tailwind and a lightweight motion layer for the timer coach and panels. web speech api handles low latency transcription and filler detection. a custom 2d barista canvas keeps the scene alive while the user talks. elevenlabs provides voice responses and gemini scores clarity, pace, and confidence once the timer ends.
speech recognition behaved differently across chrome and safari, so we added careful timing and fallback handling. keeping the timer, simulator, transcript panel, and audio playback smooth at once took real tuning. the feedback had to motivate without sounding like a grade.
pairing web speech api with custom hooks and refs made truly live feedback possible. small timing changes in animation and audio shifted the whole feel of coaching.
we want scenario switching for interviews and investor pitches, plus a personal dashboard for streaks and filler trends. longer term, practice rooms so friends can compare takes and encourage each other.