How can someone explain a customer-care problem when the helpline doesn’t offer their language?
We explored this through Pratinidhi, our prototype for the Great Indian AI Internship Challenge.
The demonstration follows this journey:
- A customer explains her concern in Marathi.
- Pratinidhi identifies missing details and asks for them by voice.
- She gives permission before the case is shared.
- The representative handles English support while she follows the conversation in Marathi.
- When additional information is needed, it asks her privately, confirms the answer and continues. We used Gnani Prisma for speech recognition and Timbre for speech synthesis.
The current demonstration combines recorded, genuine speech API outputs with a simulated support call. It explores the interaction design before live telephony integration.
Which failure mode would you prioritise before taking this into real calls?
