What it does

  • Customer care in every dialect: An assistant on WhatsApp, the app and web for plans, bills and common faults, which answers from approved sources and hands over to a person when needed.
  • Contact-centre agent assist: Suggesting the answer and next step to the agent during the conversation, from the internal knowledge base.
  • Call summaries and complaint classification: Turning Arabic calls into text, summarising them and classifying the reason, for clearer reporting and faster routing.
  • Operations runbook search: Answers for network and operations teams from internal runbooks and procedures, citing the source.

Why inside your premises

Subscriber data and call content are sensitive, and at your volumes per-call cloud pricing is hard to predict. When the model runs inside your network, the data stays with you and the cost is known in advance, because it is infrastructure you size. More in what an on-premises LLM is.

What we deploy

NoorAI deploys a complete stack that you own: Orchestra, our 14-billion-parameter Arabic-native language model trained in-house, a low-code agent platform wired into your systems, and the engineers who deploy and run it with you. It runs inside your data centre, air-gapped, or in an in-country private cloud.

How we start

We usually suggest an operator starts with contact-centre agent assist, where the agent stays in control. A typical first project:

  1. Suggestions to the agent during the conversation for one team, from the internal knowledge base.
  2. Sizing from your actual volumes: conversations at peak hour, languages and dialects, and the response time needed.
  3. Pass criteria agreed before anything is bought: the share of suggestions agents accept, response time, and a correct source.
  4. A parallel run, then the subscriber-facing assistant on WhatsApp and the app.

Then we deploy and operate it with you, with monthly support. We work in Egypt, Saudi Arabia, the UAE and across the Gulf.

Frequently asked questions

Can it handle an operator’s volumes?

We size it with you in the Discover phase from your actual conversation volumes, and agree performance criteria before anything is bought.

Does it handle messages that mix Arabic and English?

Yes. Arabic is its anchor language, and English is supported for mixed messages.

Does subscriber data leave the network?

No. The model, indexes and logs run inside your environment.