Restaurant AI Agent Platform

The problem

A restaurant assistant needs to handle several kinds of requests: business questions, menu discovery, ordering, order status, and requests for a member of staff. Each requires different information and actions, while the conversation needs to retain context.

I built Restaurant AI Agent Platform as a multi-tenant customer engagement and restaurant management platform, with assistants for websites and Telegram and support for multilingual text and voice.

Routing a conversation

An LLM-based intent classifier routes messages to the appropriate workflow. Persistent chat history and session management retain context across turns.

The system connects business knowledge retrieval, menu discovery, cart management, ordering, tracking, and staff support through the same conversation infrastructure.

Business-specific retrieval

Retrieval-Augmented Generation uses business-specific knowledge and FAISS vector search. Semantic caching reduces repeated AI processing for similar requests.

This retrieval layer sits alongside operational tools. The ordering workflow covers cart management, order placement, and Petpooja POS integration.

Staff operations

The management dashboard provides conversation history, live chat monitoring, and human-agent takeover. Staff can continue a customer conversation when direct assistance is needed.

Natural-language sales analytics provide another interface to restaurant data from the management dashboard.

Backend and deployment

The platform uses Python, Django, and Django REST Framework, with PostgreSQL, MongoDB, Redis, and Celery. Celery and Redis support asynchronous message processing. Deployment configuration uses Docker and Nginx.

My work covered conversation orchestration, retrieval, ordering integration, management dashboards, and backend infrastructure.

Demonstration

See the product overview for the customer workflow, or get in touch to discuss a demonstration and integration requirements.