AI solutions for businesses in Kurdistan and Iraq

Practical AI, inside the software you already use

Enzziro is a software company in Sulaymaniyah (Slemani) that adds AI to business software for companies in the Kurdistan Region and across Iraq. Our focus is practical: repetitive, text-heavy tasks where a language model can save people time, with a person still deciding anything that matters.

That can mean support triage, document processing, internal search, drafting, or a forecasting prototype, depending on the data you have and what the work demands. We'll also tell you when a plain rule or a better form would do the job with no AI at all.

What we build

• AI assistants and agents. Guided tools that look up information in your systems, draft replies or documents, and take actions a person has approved.
• Document and language processing. Classifying, summarising and searching documents, and pulling agreed fields out of forms, contracts or applications for human review.
• Internal search. Staff ask questions in plain language and get answers drawn from your own documents, with the source shown.
• Workflow automation. Connecting systems so routine steps happen without someone copying data between them.
• Forecasting prototypes. Testing whether your historical data can support useful signals about sales, demand or churn before anything is built on them.

Arabic, Kurdish and English

Language support is where AI projects in this region succeed or fail, so we test for it first:

• Arabic is well supported by the major language models.
• Sorani Kurdish is supported far less consistently. Quality varies by model and by task, and it is weaker for anything specialised.
• Mixed-language text, such as Kurdish written in Latin letters, Arabic with English terms, or a WhatsApp message in three scripts, needs testing on real examples.

So before we promise a result, we run the task on a sample of your real messages or documents in each language and show you where it works and where it doesn't.

Human in control

AI systems need clear boundaries. On every project we agree:

• What data goes where. What is sent to which provider, what is stored, what is logged, and for how long.
• Who can see what. Role-based access for people and for automated steps.
• When a person decides. Approvals on anything with a financial, legal or customer-facing consequence.
• How quality is checked. An evaluation set of real examples, run again whenever the model or prompt changes.
• A record of what happened. Inputs, outputs, approvals and actions logged where the workflow needs an audit trail.

Hosted, private-cloud and self-hosted options are compared for each project based on its security needs. We don't assume a privacy or security outcome before those requirements are defined.

Our work so far

Our AI work to date includes an in-house concept, an AI assistant inside a messaging app, which combines personal messaging with an assistant for instant support and recommendations, and lets users choose which AI model answers them. It's a concept project, not client work, and we say so. For client software, we add AI features to the systems we build, such as custom back offices and web platforms, where there's a repetitive task worth automating.

Technology we use

• Python and TypeScript back ends, with LangChain where orchestration helps.
• Hosted language models from major providers, or open models where data must stay private.
• Retrieval over your documents, with PostgreSQL and vector search.
• Evaluation harnesses built from your own examples.

How a project runs

1. Pick one task. We choose a task that is frequent, measurable and low-risk to get wrong while it is being tested.
2. Test on real data. A short proof of concept on your own examples, in each language, with the results shown to you.
3. Build it into the workflow, with approvals, access control and logging.
4. Measure and widen. We expand to more users or tasks only when the numbers hold up.

What affects the cost

• The quality and shape of your data. Clean, structured records are quicker to work with than scanned paper.
• Languages, especially if Kurdish accuracy matters.
• Model usage. Hosted models charge per use, so volume affects running costs as well as build cost.
• Integration and controls: approvals, audit logs and connections to your existing systems.

Describe the task on the start a project page, and we'll tell you whether AI is the right tool for it.

Frequently asked questions

Can AI understand Kurdish?

Partly. Arabic and English are handled well by current models. Sorani Kurdish support is improving but uneven, so we test your actual content before relying on it.

Is our data used to train someone else's model?

Not if we can help it. Provider terms differ, so we choose providers and settings that don't train on your data, and we document exactly what is sent. Where that isn't enough, open models can run on infrastructure you control.

Will AI replace our staff?

That isn't how we design it. The aim is to take repetitive handling off people so they spend their time on the cases that need judgement, with a person approving anything that matters.

Do we need a lot of data?

Not for most assistants and document tasks, which work from your existing documents. Forecasting is different: it needs enough clean history, and the prototype stage tells you whether you have it.