Built for Vyshak's mother, this WhatsApp agent turns natural Kannada, English or "Kanglish" voice notes into Swiggy grocery and food orders. Gnani AI's PRISMA speech-to-text and TIMBRE text-to-speech models handle Indic voice; Gemini interprets the request, Swiggy MCP searches, selects and orders, and the agent replies with a Kannada voice note. It replaces complex app flows with speech in a familiar chat. Featured in Prajavani, one of Karnataka's leading newspapers.
I built this first for my mom. Ordering groceries meant opening different apps, searching, choosing brands, adding items and checking out. WhatsApp voice notes were already familiar to her, so I asked: what if she could just say what she needed there?
The result is a WhatsApp agent for natural Kannada, English or mixed "Kanglish" voice notes. Gnani AI's PRISMA speech-to-text turns the note into text; Gemini interprets the request; Swiggy MCP can search, select, add to cart and complete the order; and Gnani's TIMBRE text-to-speech sends a Kannada voice reply. I also tested it on food: I asked for biryani by voice, and a few minutes later it was on the way. Zero text or clicks.
Prajavani, one of Karnataka's leading newspapers, featured the story as a small example of a bigger shift: instead of teaching millions of people to navigate more apps, agents can meet them in a familiar chat and in their own language.
This is still an early prototype, not a public product. The article also points to the current limit: among 100 major Indian consumer internet companies reviewed, only eight had an official first-party MCP server.
I built this first for my mom. Ordering groceries meant opening different apps, searching, choosing brands, adding items and checking out. WhatsApp voice notes were already familiar to her, so I asked: what if she could just say what she needed there?
The result is a WhatsApp agent for natural Kannada, English or mixed "Kanglish" voice notes. Gnani AI's PRISMA speech-to-text turns the note into text; Gemini interprets the request; Swiggy MCP can search, select, add to cart and complete the order; and Gnani's TIMBRE text-to-speech sends a Kannada voice reply. I also tested it on food: I asked for biryani by voice, and a few minutes later it was on the way. Zero text or clicks.
Prajavani, one of Karnataka's leading newspapers, featured the story as a small example of a bigger shift: instead of teaching millions of people to navigate more apps, agents can meet them in a familiar chat and in their own language.
This is still an early prototype, not a public product. The article also points to the current limit: among 100 major Indian consumer internet companies reviewed, only eight had an official first-party MCP server.