Helpers on call
Retail journey
Speak in your language.
Mohan, 58 · retired teacher · Vijayawada · reads Telugu, types slowly
Mohan does not want to type on a phone keyboard. Mukodi listens to him in Telugu and answers in Telugu — by voice, in text, or both.
Marketing claimeight-languages
The story · 4 scenes
Beginning to end.
Mohan, 58 · retired teacher · Vijayawada · reads Telugu, types slowly
·Mohan does not want to type on a phone keyboard. Mukodi listens to him in Telugu and answers in Telugu — by voice, in text, or both.- Scene 01 · Pick Telugu, end-to-endMohan switches the app language to Telugu. The UI, the voice prompts, and the synthesized replies all switch together — no half-translated screens.
- Scene 02 · Ask a question by voiceHe taps the mic and asks, in Telugu, how much he has spent on medical bills this year. The app transcribes the question, classifies the intent, and answers — in Telugu, by voice and by text.
- Scene 03 · Hear the answer with the reasonsThe reply lists his three largest medical merchants and the months they fell on. Each row is read aloud with the merchant name in the original script.
- Scene 04 · Switch to Hindi mid-conversationHis daughter joins the conversation and switches to Hindi. The next answer is in Hindi without losing the thread of the conversation.
Helpers on call
The stack · helpers + models
Who's actually doing the work.
Models invoked
Honest limits
What this flow doesn't do.
- Indic STT confidence is best on common phrasing; very colloquial or accented speech may need to be repeated once.
- Voice playback uses IndicTTS; long English-script merchant names are pronounced phonetically rather than translated.
- Voice mode is opt-in under DPDP §6 — audio is processed on session and is not retained beyond the answer turn unless the user explicitly saves it.