joya.

Chapter 02 · Asaf Lupo

Marketing and AI for a construction atelier

A.Lupo Group is a boutique construction and development company working in Herzliya, Tel Aviv and Ramat HaSharon. It began by building private villas, and today also develops residential buildings and urban renewal projects. Its signature is exposed concrete with a rich, unusual finish. Since June 2025, I've been its AI Systems & Marketing Lead.

Four brand films, made with AI

Project 1 · AI films

project film · Eli Cohen 12-14, Herzliya

4films for a construction atelier
0shoot days

An animated company presentation

Project 2 · Presentation

Animated company presentation · The Portfolio, in Motion

The apartment sales CRM

Project 3 · Live demo
Asaf Lupo sales CRM
Loading the live demo

The real system, screen for screen, with the three projects now in marketing, their real names and photos from the public company website, and invented leads. The apartment numbers are blurred. Used every day by the sales team. Click to open the live demo

At a glance

Projects synced from the company website

The project list comes from the website through its WordPress API and is synced daily

Project details that fill themselves

Linking a project to the catalog fills name, city and architect and brings its photo from the project page

Every lead has an owner

A new lead belongs to whoever created it, and every note records its author

Today's follow ups in one list

The home screen shows every open reminder due today or earlier, labeled later today, passed today or overdue

Late callbacks flag themselves

A lead whose reminder time has passed gets a "לחזור ללקוח" (call back) chip in the leads list

Sold apartments count themselves

Each project shows the apartments in the building, the owners', for sale and sold, with sold counted from closed deals

This month's closed deals in one figure

The top of the home screen sums the deals closed this month

Stack: Next.js · Supabase · Vercel · WordPress REST API · Built with Claude Code

3projects in marketing, each with its own leads

A fleet platform that reminds its drivers

Project 4 · Live demo
Company fleet dashboard
Loading the live demo

The real system, screen for screen, running on invented vehicles and drivers. Click to open the live demo

At a glance

Type a plate, get the car

A plate lookup against the government open data registry fills model, year and fuel type

Urgency you can see

Every date is tinted red when expired, orange within 7 days and yellow within 30, with the most urgent vehicles first

Test dates checked at the source

One click checks the test against the Ministry of Transport open registry and updates the date

Automatic morning alerts

Every morning the system checks insurance, test, license and lease-end dates and unpaid fines, and emails each driver about their own vehicle

One summary for the office

The same check sends the office a single summary

Fines follow the driver

An unpaid fine keeps alerting after a vehicle is deactivated, addressed to whoever drove it on the day of the fine

Unpaid fines, totaled

The fines tab totals what each vehicle still owes, and marking a fine paid brings the total down

Fuel watch Agent

An AI agent that reads Sonol's monthly invoice from the inbox, matches every fill-up to its vehicle and driver, and checks that the lines add up to the charge. It flags four things worth checking: higher spend than the vehicle's previous months, two fill-ups close together, a vehicle with no fill-ups all month and a km reading typed wrong. The fleet manager gets the full report, and a flagged driver's manager gets a short email about that driver

Stack: Next.js · Supabase · Vercel · Microsoft Graph (email) · government open data (data.gov.il) · Built with Claude Code

The email drivers get every morning

Every morning the system emails each driver about the items on their vehicle that need attention. This is that email, in its real wording, with invented names.

Loading

5kinds of alerts: test, insurance, license, lease end, fines
In build

An AI agent for company documents and Priority

The problem

The answers live in thousands of documents and in Priority, and finding one takes hours

What I'm building

One agent that reads both

How it will work

Ask on WhatsApp or on the desktop, and get the answer in seconds, with its source

For example: "How much have we paid the contractor on this project so far?"

Next chapter · 03 ClimateCoating

Contact

Let's talk