Multi-tenant hotel assistant with Indian-language voice
Guest requests for booking, room service, food, feedback and checkout handled by agents in more than one language.
What the client needed
Guest requests across booking, room service, food ordering, feedback and checkout, in more than one language, were handled by stretched front-desk staff.
What runs now
Each hotel group runs as its own tenant with its own menus, rooms and policies.
The system, step by step
Agents on LangGraph and LangChain, with PostgreSQL checkpointing and Redis session management
gRPC microservices and WebSocket connection management
Sarvam AI for Indian-language speech to text and text to speech
Intent classification benchmarked two ways, LLM against retrieval with Qdrant and Nomic embeddings, and evaluated with DeepEval
An outbound voice sales agent on Twilio, Deepgram and Gemini text to speech
Have a problem you want measured?
Tell us what you want to count, check or protect. An engineer will reply, not a sales script.