AI solutions

We build AI that works on your own documents and data: assistants that answer with the source they used, AI agents that handle bounded tasks and pass the result to a person for approval, and tools that read photos and scans into structured fields. A project can start with a proof of concept on a sample of your real material, so you see how the system behaves before you commit to a full build.

Who it is for

This service fits teams that lose time searching long documents, retyping details from photos and scans, or answering the same questions every day. It also fits a company adding a practical AI feature to a web or mobile product it already runs, and an organisation that wants AI on confidential material with a say in where that material is processed.

If a fixed rule or a simple database query already gives the right answer every time, you do not need AI. We will tell you so on the call and point you to automation or a plain backend instead.

What we build

• Assistants on your documents. Ask in Arabic or English and get an answer drawn from your manuals, policies and contracts, with the document and page it came from. When the answer is not in your documents, the assistant is designed to say so rather than guess, and the test set before launch checks how well it does.
• AI agents for bounded tasks. Agents that sort incoming messages, pull the key details out of messy requests or invoices, check an order or booking in your system, and draft a reply for a person to approve before it goes out.
• Document and image reading. Computer vision pipelines that read receipts, ID cards and order slips in Arabic and English, fill the matching fields, and mark the uncertain ones for a person to check.
• Graph RAG and machine learning. Graph RAG, retrieval-augmented generation over linked data, connects orders, stages, suppliers or logs, so a manager can ask which earlier step is linked to a delay and see the records behind the answer; a person still confirms the root cause. Where your data supports it, we also train models that forecast demand for your products or delivery times and flag unusual records.

How we build it

You choose the model before we build: a commercial model through a paid business API, or an open-weight model that runs on servers you control. We go through the provider’s data terms with you before any of your data is sent; whether they meet your obligations is for your legal adviser. Where you use an external model, we can mask personal details such as ID and phone numbers before text leaves your systems. Where processing happens depends on the provider and region you pick, and we write it down for each project.

We apply your access rules at the search step, before the model sees any text, so answers are drawn from the documents the user’s role may read. Before launch, we build a test set of real questions with your subject expert, including questions the assistant should decline, and measure whether it finds the right passage and stays faithful to it. You get the results in writing. We test on the Arabic your users actually write, including Egyptian or Gulf dialect where it matters.

We work in Python and FastAPI or TypeScript and NestJS, with PostgreSQL and pgvector or Qdrant for search, Neo4j for linked data, OpenCV for images, Ollama or vLLM for self-hosted models, and n8n when an agent needs to act in your other tools.

Timeline and cost

Time and cost depend mostly on:

• Your source material. Clean digital files are quicker to prepare than scanned pages with tables, stamps and handwriting.
• What the system does. An assistant that only answers questions is smaller than an agent that works with several live systems or with linked data.
• Where the model runs. A cloud API needs no hardware on your side; a self-hosted model needs prepared servers and tuning on them.
• How strict the evaluation is. The size of the test set and the rounds it takes to reach your accuracy target.
• What it connects to. Your app, a chat channel or an ERP.

Model usage and hosting are running costs, separate from the build, and the scope estimates them so you can plan.

After the free call, you get a written scope with the timeline and the price, before you pay any deposit.