AI automation & chatbots

Five kinds of AI work we ship, each scoped as one job with a measurable outcome.

Where we have shipped AI, and where the same patterns apply.

• A one-week discovery that picks one job and defines how success is measured
• Connection to your documents, database or APIs so answers are grounded in your data
• The chat or workflow interface inside your website, app or WhatsApp
• Guardrails: approved sources only, refusal when unsure, handoff to a human, full logs
• An evaluation set of real questions run before launch and after every change
• Arabic and English, with French on request
• Monthly usage caps and cost reporting
• Handover, monitoring and tuning after rollout

1. Discovery. One week. We pick the job, the data sources and the success metric.
2. Prototype. A working version within three weeks, tested on your real questions.
3. Pilot. Real users, a review of transcripts, guardrails tightened.
4. Rollout. Live in your product with caps, logging and a human fallback.
5. Tune. Monthly review of usage, cost and misses; content and prompts updated.

We quote per project, within one working day, once the scope is clear. These are the variables that move it.

• Number of jobs, channels and languages
• How clean and accessible your data sources are
• Integrations: CRM, WhatsApp Business, POS, existing CMS
• Model usage volume, passed through at cost
• Whether it lives in a platform we built or one we take over
• Monitoring and tuning retainer after rollout

Tell us the one job you want automated and where the data lives. We reply within one working day and propose a one-week discovery with a prototype three weeks later.

Model provider chosen per job, not tied to one vendor · Retrieval over your documents · Next.js and TypeScript · Supabase or Firebase · WhatsApp Business API

Built in Byblos, Lebanon. Serving Beirut, the Gulf and clients abroad remotely.

AI that does a specific job inside your product. Chatbots that answer from your documents, tutors that adapt to each student, intake flows that turn forms into structured quotes. Built in Byblos, Lebanon, in Arabic and English, with your data staying in your tenant.

AI automation means using a language model to do a defined task inside a business process, such as answering customer questions from a document set, grading a student's answer, or reading an inbound form and drafting a quote, with a person reviewing anything the model is unsure about. It is not a chat window bolted onto a website. We scope one job, connect the model to your data and your existing web platform, test it against real examples, and cap what it can cost per month.

Questa Education runs on an AI engine we built to identify each learner's weak points, generate personalised plans and produce assessments automatically, across three portals. The same team teaches the MoonDev × Ranna Studios AI Program in Lebanon, twelve 90-minute sessions across five modules for people with no coding background, so we understand the tools from both sides: building them and explaining them.

Every deployment ships with guardrails. Your data stays in your tenant. The model answers only from the sources you approve and falls back to a human when it is unsure. Monthly usage is capped so costs stay predictable. Arabic and English out of the box, French on request. We start with a one-week discovery and a working prototype within three weeks.

• 360° Virtual Tours: Let buyers and guests walk through your space from any phone, before they arrive.
• Augmented Reality (WebAR) & AR Menus: Dishes, products and business cards in AR from a QR code. Nothing to install.
• Web & Mobile App Development: Next.js websites, installable web apps and the CMS behind them. You own the code.

The AI Program we teach · All five services · Browse all 35 projects

We choose the model per job and you can set constraints, for example a provider that is approved by your IT team or a region where data must stay. Most work uses a hosted model for language and a retrieval layer over your own documents; where the job is narrow, a smaller or open-weight model runs cheaper. We are not tied to one vendor and we tell you why we picked one.

In your tenant: your database, your storage and your accounts, in Lebanon or the region your policy requires. The model only receives the passages it needs to answer a question, and we can restrict providers to those that do not train on your data. Nothing is copied into a MoonDev account.

Yes. We index your PDFs, pages, spreadsheets or database and the assistant answers only from those sources, citing which one it used. When the answer is not in the sources it says so and offers a human, instead of inventing one. Updating a document updates the answers.

Four ways. Answers are grounded in approved sources only. The assistant is instructed to refuse when it is not sure and hand off to a person. We run an evaluation set of real questions before launch and after every change. And every conversation is logged so misses are found and fixed, not guessed at.

A build fee for the discovery, prototype and rollout, then monthly model usage passed through at cost with a cap you set, plus an optional tuning retainer. Usage for a support assistant on a Lebanese SME site is small compared with the build; the cap keeps it predictable. We do not publish prices here; we quote after the one-week discovery.

Yes. Every assistant we ship handles Arabic and English, including Lebanese dialect in WhatsApp conversations, and French on request. Your documents can be in one language and the answer in another. We test both languages in the evaluation set before launch.

Yes. Most of our AI work goes into platforms that already exist, through an embedded widget, an API or a WhatsApp number. We take a short look at the stack first and tell you whether it slots in or needs a small bridge. If we built the platform, as with Questa and Edu4All, the integration is direct.

Yes, when it is connected to the system that holds them: a reservation tool, a delivery platform or a CRM. The assistant collects the details, confirms them with the customer and creates the booking or order, and a person is notified for anything outside the rules. We start with answering and add actions once the answers are reliable.