MuhammadSobanHameed

Whatsapp AI Ordering Agent — AI skill for Claude Code

AI community

Case study: a production multi-tenant WhatsApp AI ordering agent built on Claude tool use, with deterministic guardrails for prices, totals and order placement.

How to install Whatsapp AI Ordering Agent

This entry records only its repository, not the path inside it, so there is no exact command to give. Open MuhammadSobanHameed/whatsapp-ai-ordering-agent and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Whatsapp AI Ordering Agent does

Case study: a production multi-tenant WhatsApp AI ordering agent built on Claude tool use, with deterministic guardrails for prices, totals and order placement.

Alternatives in AI

CliRelay — Self-hosted AI gateway for coding CLIs — one OpenAI/Claude/Gemini/Codex-compatible endpoint, with a multi-tena 1k ★
Style Distill — style-distill skill — 三阶段 13 模板 LLM 蒸馏（style-distill-order） 669 ★
Claude Code 2.1.88 — Source code of @anthropic-ai/claude-code v2.1.88, extracted from the npm package's source map (cli.js.map) 169 ★

README

WhatsApp AI Ordering Agent — Case Study

A production, multi-tenant WhatsApp bot that lets customers browse a shop's catalog, build a cart, and place real orders by chatting naturally in English, Roman Urdu or Urdu. It handles text and voice notes. Built on Claude (Anthropic) with tool use, and wrapped in deterministic guardrails so the AI can never invent a product, misquote a price or place an order the customer didn't agree to.

**Author:** Muhammad Soban Hameed · **Role:** AI / Backend Engineer (design, implementation, production support) **Status:** Live in production, serving multiple businesses (restaurants, pharmacies and retail shops)

[!NOTE] **This repository is a case study, not the source code.** The code samples are **simplified illustrations written for this document**, not excerpts from the real codebase. Company names, client names, endpoints, credentials and customer data are deliberately left out.

The problem

Small businesses already take orders on WhatsApp, but by hand. Someone has to answer every "menu bhej dein", work out the prices, take down the address and type the order into the ERP. Messages arrive at all hours, often misspelt, mixed-language or as voice notes.

The goal was a bot that does this end to end **and writes real orders straight into the company's ERP**. A wrong price or a phantom order costs the shop real money, so "mostly right" wasn't good enough.

What I built

Area	Highlights
Conversational ordering	Browse categories, search items, add, remove or change quantities ("one more", "make it 3"), choose add-ons, delivery details, special instructions, order summary, and placement on a plain "yes"
AI agent	Claude with 15 tools, running in a tool-use loop. The model decides intent, and code owns money, cart state and order placement
Guardrails	Deterministic validators catch invented products or sizes, misquoted prices, wrong totals, false "order place