Hinglish Coach

Inspiration

Most language mastery equipment grades speech towards a proper, textbook trendy — but that is really no longer how bilingual Indians truly speak. I paint in real-time voice AI (I run a voice-calling AI company), and the single biggest salient signal in a spoken Hindi-English communique isn’t grammar — it’s whether or not the code-switching between languages sounds herbal or forced. No current device trainer for that particular. So I built one.

What it does

Hinglish Coach listens to a sentence you communicate out loud - in some natural blend of Hindi and English you'll actually use - and offers you spoken, demonstration-style feedback: become code-switching herbs, modified to write up by (too formal "textbook" Hindi no one could accommodate anymore), and whatever and. The feedback is checking the lower back to you right now, so the whole loop is voice-in, voice-out.

How We Built It

An unmarried-page vanilla HTML/CSS/JS frontend uses the browser's local Web Speech API for each speech-to-text and text-to-speech — no external transcription or voice services desired, so it's unattached and instantaneous.
A small Node.Js + Express backend with an endpoint that sends the transcript to the OpenAI API with a coaching-accurate tool set off tuned for speech Hindi-English code switching, now not textbook grammar correction.
Feedback is usually lower back in English so it is understandable no matter who is listening, even when the input changed to Hindi or Hinglish.

Challenges We Faced

Getting the exercise set off correctly changed to the hard part — early variations gradually sentences like a grammar checker, that's exactly the wrong instinct right here. Set off had to be adjusted to praise natural code switching in favor of punishing it, it’s a subtler judgment than it sounds.