Document extraction
- Goes in
- Invoices, forms, ID scans, PDFs, photos of paper
- Comes out
- Structured fields in your system, with a confidence flag
- Who checks
- A person reviews anything below the confidence bar
AI solutions
AI that does one job well, inside your operation.
Not a chatbot bolted on. Assistants, extraction, classification and search built into the systems you already run, with a person in control at the steps that matter, in Arabic and English.
Try the shape of it
Paste a message or an invoice. See what comes back.
A browser simulation with simple rules, not a model. Real projects use a model tested on your own examples; the output shape and the human gate are the same.
Try:
Type or pick a sample to see category, priority, owner, confidence and a drafted reply.
Runs entirely in your browser. Nothing you type here leaves the page.
What we build with AI
Six jobs, each with a clear input and output.
Pick the one closest to your problem. Every card says what goes in, what comes out, and who checks it.
Request classification and routing
- Goes in
- WhatsApp, email and form messages
- Comes out
- A category, a priority, the right owner and a drafted reply
- Who checks
- Your team sends or edits the reply; nothing goes out on its own
Assistant on your own documents
- Goes in
- Your policies, manuals, catalogues and records
- Comes out
- Answers with the source shown, for staff or customers
- Who checks
- Answers cite the document; unknown means unknown, not a guess
Search across records
- Goes in
- Customers, cases, files scattered over systems
- Comes out
- One search box that understands meaning, not just keywords
- Who checks
- Access follows your existing permissions
Summaries and drafts
- Goes in
- Long threads, reports, meeting notes
- Comes out
- Short summaries, first drafts, translations between Arabic and English
- Who checks
- Always a draft; a person signs it off
Arabic-first assistants
- Goes in
- Arabic dialects, mixed Arabic and English, right-to-left forms
- Comes out
- Assistants and extraction that handle Arabic as a first language
- Who checks
- Tested on your real examples before launch
A person in control
A person stays in control. Your data stays yours.
- 01Start with real examples
We ask for twenty real documents or messages before we promise anything, and test against them first.
- 02Confidence, not confidence tricks
Every automated result carries a confidence flag. Low confidence goes to a person, never straight through.
- 03Show the source
Assistants quote the document they used. If there is no source, the answer says it does not know.
- 04Small first, then wider
One workflow, one team, measured for a few weeks, then extended if it earned it.
- Your data stays in your systems. AI steps read what they need for the task and write results back to you.
- Your data is used to answer your requests, not to train anyone's models, unless you decide otherwise in writing.
- The model provider is chosen per project and named in the plan, together with where it runs and what it is allowed to see.
- Health, school and pharmacy data get the strictest option available for the project, agreed before the first line of code.
Already running
Two of our own products use it every day.
Questions about AI
What people ask before an AI project.
Which model do you use?
It depends on the job. The plan for your project names the model and provider, where it runs and what it may see. This site's own assistant runs on Cloudflare Workers AI (Llama 3.3 70B).
Is my data used to train models?
No. Your data is used to answer your requests, not to train anyone's models, unless you agree otherwise in writing.
How good is the Arabic?
We design the flows Arabic-first, including dialects and mixed Arabic and English, and test them on your real examples before launch. Two of our own products run in Arabic every day.
What does a first AI project look like?
One job in one workflow, shipped in reviewable increments. The first conversation is free and ends with a written starting point.
Have a document pile or a message queue?
Bring twenty real examples. We will tell you honestly what AI can and cannot do with them.