tanhAI: A Conversational Food Agent led by Jaggu
Turning chats and saved reels into real orders across Swiggy food, Instamart, and Dineout, guided by Jaggu, a monkey who speaks 11+ languages.
We live in an age of infinite scrolling. You see a mouth-watering food reel, save it to a collection you will never open, and move on. tanhAI closes that gap: describe a dish, paste a reel, or simply chat, and Jaggu, a playful monkey who speaks 11+ languages, turns the craving into a real order at your doorstep.
Under the hood, tanhAI runs on Sarvam's model stack: a 30B model handles intent parsing while a 105B model drives the main reasoning, and the Saaras model powers the speech layer so Jaggu can talk back across languages. Swiggy's MCP wires the agent into food delivery, Instamart groceries, and Dineout, fetching real-time stock and localized pricing and populating carts, all without ever opening the app.
Key Contributions
- Jaggu, the Multilingual Monkey: The heart of the app, a playful monkey persona who speaks 11+ languages, making ordering feel like chatting with a friend rather than tapping through a delivery app.
- Sarvam Model Stack: A Sarvam 30B model handles fast intent parsing, a Sarvam 105B model serves as the main reasoning model, and the Saaras model powers the speech layer that lets Jaggu listen and talk back.
- Swiggy Model Context Protocol (MCP) Integration: Connected the agent directly to Swiggy across food delivery, Instamart, and Dineout, performing real-time stock-checking, fetching localized pricing, and populating carts without ever loading the app UI.
- Craves Reel Pipeline: Paste a short reel and tanhAI extracts and transcribes its audio (no vision model, it's purely audio-driven) to pull out dishes, cuisines, and restaurants, then cross-references local availability. It also nudges mindful choices: save a gym reel earlier and reach for a burger later, and Jaggu gently flags it.
