Portfolio build — A production-grade ordering platform built end-to-end by Khoda Consulting, paired with a separate live AI agent demo. Built to show capability — not a paid client engagement. Want something built like this?

Portfolio Build · Live Demo

Full-Stack Commerce

AI Build & Implementation

Delivr + FreshHub Food Delivery Platform

A complete, custom technology stack for a multi-restaurant hospitality scenario — a multi-tenant ordering platform with real Stripe payments and operations tooling (FreshHub), paired with a separate interactive AI agent demo (Delivr) that shows how an agent layer would handle support, menu intelligence, and personalization. Built end-to-end by Khoda Consulting. Zero off-the-shelf SaaS.

Portfolio Build · Khoda Consulting Khoda Consulting Next.js 15 · Supabase · Stripe React · Node.js · Claude API Vercel

Try the live demo Work with us

100%

Custom-built, owned code

2×

Layers — Commerce + AI

Off-the-shelf SaaS used

Agent tools in the live demo

The Scenario

Build the entire technology stack from scratch

Khoda Consulting built FreshHub to model a common problem end-to-end: a multi-restaurant hospitality operation that needs to modernize without generic SaaS tools — a fully custom, owned technology stack that can grow with the business and treat AI as a first-class capability. The platform is real and production-grade; the operator scenario below is the use case it was built to solve.

No existing ordering infrastructure — customers were calling in orders or using expensive third-party platforms that took 20–30% commission.

Multiple restaurant locations with different menus, hours, and pricing — needed a single admin system that owners could manage from their phones.

Restaurant owners lacked time to manage menus, write descriptions, respond to customers, or track onboarding progress without AI assistance.

Speed and ownership mattered — the goal was a production-ready platform that could go live and take real payments without months of work, with the architecture designed to scale without redeploys.

The Build

Two builds — a platform and an agent.

Rather than stitching together existing tools, Khoda Consulting built both pieces from scratch: a production-grade commerce platform (FreshHub) as the foundation, and a separate interactive AI agent (Delivr) that demonstrates how an operations layer would work. They are independent builds — the agent demo is not wired into the live platform.

Layer 1 — Commerce Platform

FreshHub

Multi-tenant ordering platform — custom menus, Stripe payment processing, role-based admin, real-time order management, Google Places onboarding, and scheduled pickup with hours validation. Built end-to-end and production-ready; not yet live, and contains no AI layer.

Layer 2 — AI Agent

Delivr AI

Multi-tool AI agent built on Claude — autonomous customer support, menu description generation, review response drafting, onboarding guidance, and personalized recommendations. A separate, live interactive demo running on simulated order data to show how an agent layer would operate; it is not integrated into the platform.

Architecture

How the full stack is built

The platform runs on a 7-layer architecture spanning the commerce foundation and AI layer — from customer ordering through payment processing, operations management, and autonomous AI tooling. Each layer is independently deployable and replaceable.

Unified Platform — System Architecture

C1Customer Interface — Next.js 15 App Router

Restaurant menu pagesCart + checkoutStripe PaymentElementMobile responsiveLanding page

↓ Order saved → /api/create-order (service role)

C2Commerce Layer — Supabase + Stripe

Multi-tenant schemaStripe PaymentIntentsRLS + service roleJSONB order storageResend email

↓ Admin operations — role-based access control

C3Operations Layer — Admin Panel

super_admin + owner rolesMenu CRUD + modifiersOrder status flowPromo managementGoogle Places onboarding

↓ AI layer reads order + menu data

A1AI Reasoning — Claude Sonnet + Tool Use API

claude-sonnet-4Tool use APIMulti-turn contextAgentic loopstop_reason: tool_use

↓ tool_use blocks → serverless execution

A2AI Tool Layer — 5 Serverless Functions

check_order_status()generate_menu_description()draft_review_response()get_onboarding_checklist()get_recommendations()

↓ all layers deployed together

I1Infrastructure — Vercel + Supabase + GitHub CI/CD

Vercel serverlessSupabase PostgreSQLService role API proxyGitHub Actions CI/CDHTTPS + env vars

What the Agent Can Do

Five tools, one agent

Each tool is defined with a JSON schema that Claude uses to reason about when and how to call it. The agent selects the right tool — or sequences multiple tools — based on the user's request.

Order Tracker

Looks up real-time order status by order ID

Returns status, ETA, driver, and items for any order ID. Eliminates the most common support request without human involvement.

Support

Menu Writer

generate_menu_description({ dish_name, ingredients, cuisine_type? })

Generates compelling dish copy from minimal input. Removes the copywriting barrier for time-strapped owners — especially non-native English speakers.

Marketing

Review Responder

Drafts a professional response to a customer review

Classifies review sentiment (positive / neutral / negative) and drafts a tone-matched response. Protects brand reputation without owner effort.

Marketing

Onboarding Status

Returns a customized onboarding checklist for new restaurant partners

Returns completion %, completed and pending steps, and the highest-impact next action. Cuts go-live time from days to hours with zero manual check-ins.

Ops

Recommender

Suggests relevant menu items based on customer history and context

Returns personalized picks based on order history, time of day, and neighborhood trends. Drives repeat orders — the core metric in food delivery economics.

Growth

Try It Yourself

Multi-turn reasoning in action

The agent handles complex multi-step requests in a single thread — calling multiple tools, feeding results back into Claude's context, and synthesizing a single coherent response.

What's the status of order FH-1042?

▸ tool_use — check_order_status

{ "order_id": "FH-1042" }

✓ tool_result

status: "Out for Delivery" · driver: "Miguel R." · eta: "~8 minutes"

Your order from Casa Verde Kitchen is on its way — driver Miguel R. is about 8 minutes out. Your Carne Asada Plate and Horchata are coming right up.

Draft a response to this 2-star review: "Cold food, arrived 45 minutes late."

▸ tool_use — draft_review_response

{ "rating": 2, "restaurant_name": "Casa Verde Kitchen", "review_text": "..." }

✓ tool_result

tone: "negative" · draft_response: "We're truly sorry..."

"We're truly sorry to hear about your experience at Casa Verde Kitchen. This is not the standard we hold ourselves to — please reach out to support@delivr.com so we can personally make this right."

The Agent in Action

How the agentic loop works

User sends a natural language request

POST /api/chat → { messages, tools, system }

No forms or menus. The agent understands intent from plain language and begins reasoning about which tool — or sequence of tools — is required.

Claude selects and calls a tool

stop_reason: "tool_use" → content[].type === "tool_use"

When Claude decides to use a tool, the API returns with stop_reason "tool_use". The frontend reads the tool name and input parameters from the response content blocks.

Tool executes server-side

executeTool(name, input) → structured JSON result

The tool function runs — querying data, generating content, or executing logic — and returns a structured JSON result that gets appended to the message history.

Result feeds back to Claude

{ role: "user", content: [{ type: "tool_result", ... }] }

The tool result is added to the conversation as a tool_result content block. Claude receives the full context and continues reasoning — calling more tools if needed.

Claude synthesizes a final response

stop_reason: "end_turn" → content[].type === "text"

When Claude has all the information it needs, it returns stop_reason "end_turn" with a natural language text response that synthesizes everything from the tool calls.

Tech Stack

Technology used

Frontend

Next.js 15

Database

Supabase (Postgres)

Payments

Stripe

AI Model

Claude Sonnet

AI API

Tool Use API

State

Zustand

Resend

Location

Google Places API

Hosting

Vercel

Auth

Supabase Auth

Security

RLS + Service Role

CI/CD

GitHub Actions

What It Demonstrates

What the build shows

Production-ready in weeks, not months. A full custom commerce platform built end-to-end and ready to take real payments — the kind of scope most agencies quote 3–6 months for.

Zero commission, full ownership. The client owns every line of code — no 20–30% platform fees, no vendor lock-in, no monthly SaaS subscriptions.

Unlimited restaurants, zero redeploys. The multi-tenant architecture means adding a new location is a database insert — no code changes, no downtime.

AI-ready from day one. The data architecture was designed for AI integration — order history, menu data, and customer behavior are all structured for model consumption.

Full operational tooling for non-technical owners. The mobile-responsive admin panel lets restaurant owners manage menus, track live orders, and update hours from any device.

About This Build

An honest note on what this is

FreshHub is a real, production-grade platform Khoda Consulting designed and built end-to-end — not a paid client engagement, and not yet live. Delivr is a separate, live AI agent demo you can try right now. We show both as proof of build capability, not as a deployed client result. When a system is a demo, we say so.

Want something built like this?

Khoda Consulting designs and ships AI agents, data pipelines, analytics tools, and ML solutions for growing businesses.

Start a conversation →