Kshitij Works

Independent AI products & engineering.
Gurugram, India.

← Engineering case studies

Generative AI / Conversational Commerce / AI Agents

Restaurant AI Agent Platform

A self-hosted, multi-tenant restaurant assistant for menu questions, basket updates and checkout, with multilingual text, voice and staff support.

Available on GitHub Role: independent builder & architect
Product overview & demo enquiries ↗
A connected conversation
A guest starts a conversation WEB / TELEGRAM Restaurant AI Agent Platform Business knowledge RETRIEVE Menu & ordering ACT Staff support HAND OVER
Workflow illustration

The problem

A restaurant assistant needs to answer questions, explain the menu, update a basket, and guide checkout. Each request needs the right information and actions, while the conversation retains context and follows the restaurant’s rules.

I built Restaurant AI Agent Platform as a self-hosted, multi-tenant assistant for cafés and restaurants. Website chat, Telegram, and voice share the same workflow, with multilingual replies.

Routing a conversation

A LangGraph workflow classifies requests, asks for missing details, and resumes unfinished tasks. Redis stores conversation state across turns.

The model proposes actions; backend handlers validate menu choices, prices, and checkout rules before making changes. Placing an order requires customer confirmation.

Restaurant AI Agent Platform — conversation, tools and persistence
Customer
Website assistant
Telegram
Voice
Conversation API
LangGraph workflow

Tools

Published knowledge
Restaurant menu
Basket / checkout
Order tracking
Payment / POS adapters
Human support

Persistence

PostgreSQL
Redis

Async

Celery + Redis

Restaurant knowledge and ordering

Answers use each restaurant’s published knowledge. Small knowledge sets are supplied whole; larger sets use keyword search. Draft edits stay separate until publication, and live prices come from the catalog.

Customers can choose variants and extras, update their basket, check out, and ask about order status. Adapters connect external menus, payments, and POS systems. Each restaurant has one checkout location; live connections need configuration and testing.

Staff operations

The dashboard provides menu and knowledge management, conversation history, live chat, and human takeover. Staff can pause the assistant and reply directly.

Natural-language queries offer another way to explore restaurant data.

Backend and deployment

Python, Django, and Django REST Framework provide the backend, with PostgreSQL, Redis, and Celery. Docker Compose runs the application behind Nginx, with a separate HTTPS production setup.

My work covered conversation orchestration, knowledge handling, checkout integrations, management dashboards, and deployment.

Demonstration

The Shop Assistant repository includes a sample café and payment/POS simulators. See the product overview or get in touch for a demonstration.

Technology

PythonDjangoDjango REST FrameworkOpenAI APILangGraphLLMsPostgreSQLRedisCeleryDockerNginx

Related

Let’s talk

Planning an AI assistant or ordering platform? Discuss your project.

From conversation workflows to checkout integrations and staff support, this is the kind of system I build.

Get in touch