// NIJITECH · ENTERPRISE SOLUTIONS
We build GDPR compliant AI systems that run in production rather than stopping at the pitch deck. Because we run the same core inside our own sixteen products, we have tried the architecture we recommend on ourselves first.
nijitech is an Istanbul-based technology company providing KVKK and GDPR compliant AI consultancy and end-to-end solution development to organisations. It works across AI assistants and agents, autonomous workflows, web and mobile products, and data integration. In every solution data is processed on-premise or in the cloud, outputs are tied to their source and critical decisions pass human approval.
01 — END-TO-END SOLUTIONS
These are not descriptions of abstract capability; they are problems our products already solve in production. The same approach can be adapted to your problem.
Decisions get made in the meeting, but who does what never gets written down. A week later nobody remembers and the same discussion happens again.
The meeting is heard in real time, transcribed with speaker separation, and what was said is split into action items with an owner and a date. Every summary item is tied to its source in the transcript, so the “I never said that” argument ends with the record.
First the wrong product gets picked, then the price and listing quality of hundreds of products cannot be tracked by hand. When a competitor cuts a price it is noticed hours later, and the Buy Box is already lost.
Before you sell, demand, competition and cost data combine into a profitability forecast; once you are selling, listings are optimised for search and prices adjust live against the competition. Every pricing decision is reported with the competitor move behind it — the automation is not a black box.
The customer writes on WhatsApp, asks on the marketplace and follows up by email. Three agents on three screens, none of them aware of the others.
All channels merge into one inbox. AI drafts the reply, an agent approves and sends it. Whichever channel the customer writes from, the same history is in view.
In operations that run in the physical world, two things break at once: matching the right vehicle with the right person, and collecting the money safely. Both are regulated, so the “let us just ship something quick” approach does not survive.
Matching, routing and live tracking run in one flow; on the payment side, instalments, escrow and chargeback protection come into play. The user is shown how the price was calculated, and every transaction lands in a legal digital archive — when an audit arrives, the record it asks for is already there.
Budget spreads across several channels, but which channel actually pays off gets lost between reports. Because the volume of decisions is higher than one person can review daily, the budget drifts towards whichever screen was looked at last.
Channel performance is measured continuously and budget shifts autonomously towards return. You can see which data each shift was based on; the automation’s decision can be reversed, and human approval is requested at critical thresholds.
A team that wants to build an on-chain product ends up wrestling with infrastructure before the actual product: separate integrations for swaps, token distribution, NFTs and staking — each with its own audit and its own fees.
Swaps (DEX), NFT infrastructure, a launchpad and staking come together on one EVM-compatible chain. Existing Ethereum tooling keeps working, transactions stay verifiable on-chain and fees stay low. The team builds its product, not its plumbing.
Server management sits in one panel, remote support in another tool and the VPN in a third place. Each tool keeps its own user list and its own log; who had access to what only gets investigated after an incident.
Virtual machines, containers, networking, backup and VPN come together in one panel, while remote desktop, SSH management and bulk software deployment run from the same place. Devices can connect with end-to-end encryption without passing through a central server. The infrastructure stays on your hardware — you decide where the data lives.
02 — AREAS OF EXPERTISE
Systems that plan their own steps towards a given goal, call tools and change direction based on intermediate results. What separates them from a model that answers one question is that they can carry a multi-step job from start to finish.
Workflows that step in where rule-based automation runs out. When the input does not arrive in the same shape every time — free text, a differently formatted document, a missing field — a rules engine stops. These do not.
The product the AI layer actually sits on. Without interface, performance and accessibility, even the best model goes unused.
For AI to work at all, the data has to be reachable. Collecting data from scattered systems, cleaning it and connecting to existing enterprise software — in most projects this is where the real work is.
03 — PROOF
What most consultancies can show you is a reference list. What we can show you is sixteen products running on the same core with real users. That is the only way to see whether an architecture holds up in production.
For all of them: the product portfolio →
04 — HOW WE WORK
Every stage ends with a concrete output. We do not move on until the previous stage has produced its output — so the “what are we actually doing here” question never comes up mid-project.
We clarify the current process, the data sources and the success measure together. The output of this step is not a proposal but a measurable problem statement: “this job takes this long, the target is that”.
Output A problem statement and a success measure
Scope, data flow and architecture are worked out. Which decisions are automatic and which pass human approval is settled here — not later.
Output Scope, architecture and approval points
We start with a narrow pilot, measure it, then widen. If the pilot fails there is no rollout; that is also a result, and we say so up front.
Output A working system and a measurement report
05 — SECTORS
In each of the areas below we have a product running in production, which means the sector knowledge comes from a working system rather than from theory.
The product decision before you sell, price and visibility while you sell. Because the volume of decisions is beyond manual tracking, the automation has to report the reasoning behind each one.
Customers write from seven different channels and expect the same history in all of them. The hard part is not drafting the reply but merging the channels into one inbox while the tone stays in human hands.
An operation that runs in the physical world: matching has to happen in seconds, the route has to be tracked live, and how the price was calculated has to be showable to the user.
A regulated area. Instalments, escrow and chargeback protection are legal requirements as much as technical ones, and every transaction has to land in a legally valid archive.
Budget spreads across five platforms and which one pays off gets lost between reports. Autonomous allocation works, but spending has to stay protected by guardrails.
A decision made in a meeting gets discussed again a week later if it never gets written down. A recording is not enough; the output needs an owner, a date and a source.
The other sectors automate a business process; this is the layer those processes run on. Server management, remote access and networking — in all three, what matters most is that who accessed what stays traceable.
06 — TIMELINES AND PRICING
We cannot put a number here because we do not know the scope. What we can put here is the logic pricing follows — so there is no surprise when you reach the proposal stage.
A price given without knowing the scope is wrong either against you or against us. That is why discovery is a separate, fixed-price stage. At the end you hold a measurable problem statement and a scope document — something you can use even if you do not continue with us.
Because the scope is known when discovery ends, development is quoted at a fixed price. Hourly billing is a model where scope stays vague and the customer carries all the risk; we do not use it.
In most projects the delay is not on the model side but in getting to the data. Who can grant access to which system, what format the data is in and whether it needs cleaning — these set the timeline. That is exactly why we surface them during discovery.
07 — WHAT WE NEED FROM YOU
We run the project, but there are three points where we need you. We write them down up front so nothing stalls later.
BEING STRAIGHT WITH YOU
We do not take on every job. In the cases below we will tell you so in the call, and neither of us loses time.
08 — DATA SECURITY AND COMPLIANCE
Enterprise buyers ask these at the start of the purchasing process, not halfway through. The answers are here; you can pass them straight to your legal team.
The European Union region is preferred for model calls. Where a transfer is required, safeguards under KVKK art. 9 apply and recipients are disclosed on request.
KVKK art. 9 · GDPR data residencyCustomer data is not used to train models. A model trained on your data does not go on to serve another customer.
Contractual commitmentEvery output is tied to its source: you can see which conversation a summary came from and which data a score derives from. Audit and appeal are only possible with that trail.
Explainable AI principleCritical decisions pass human approval before they go out. Which decisions fall into that scope is agreed together during the design stage.
Human-in-the-loopHow long data is kept is defined in the contract. Indefinite retention is not the default.
Data minimisationThe data processing addendum, the privacy policy and the KVKK notice are published and open to review.
Legal documentsLegal documents: KVKK notice · privacy policy · data processing addendum
FREQUENTLY ASKED QUESTIONS
Discovery usually takes one to two weeks and ends with a measurable problem statement. Pilot length depends on scope; the aim is for a narrowly defined pilot to produce a measurable result within a few weeks. The exact timeline depends on whether data access is ready — in most projects the delay is not in the model but in reaching the data.
The European Union region is preferred for processing. Where a transfer is required, technical and organisational safeguards under KVKK art. 9 apply and the recipients are disclosed on request. The current provider list is published in Exhibit C of the Data Processing Addendum.
No by default. There are two exceptions: in proof-of-concept work, where the participating company has given written permission; and for the development of a customer's own platform, at that customer's own request. Outside these cases customer data is not used for model training, and a model trained on one customer's data does not serve another.
Critical decisions pass human approval; the system prepares the recommendation, a person makes the call. On top of that every output is tied to its source — you can see which data a result came from. A wrong output can be challenged and its reasoning requested.
In most projects the integration layer is the real work. We build working integrations with calendar, CRM, messaging and accounting systems. Which systems you connect to — and whether those connections are technically possible — is settled during discovery.
Either is possible. If one of our products fits your need directly, we start with it. If not, we build something specific for you on the same core. That decision is one of the outputs of the discovery call.
Discovery and development are priced separately. The reason discovery is kept separate is this: a price given without knowing the scope is wrong either against you or against us. Once discovery clarifies the scope, we issue a fixed quote for development.
Size is not what decides it; having a defined problem is. Narrow projects that solve one job are usually the ones that produce results fastest. In the discovery call we look together at whether the problem can be defined that way.
The system keeps running on your side. Maintenance and further development are discussed separately; we do not aim for a setup that creates dependency — the code and the architecture are built to be handed over from the start.
With a discovery call. You do not need to arrive prepared; telling us which work is wearing you down is enough. We get to a measurable problem statement together.
For general questions see the FAQ page · for terminology see the glossary
ENTERPRISE ECOSYSTEM
For more than two years we have been building end-to-end AI and digital transformation architectures for leading public and private sector organisations in Türkiye. We do not deliver a single product and withdraw: we take on the whole technology roadmap, from analysing the existing system to architectural design, from development to integration.
AREA 01 · DEFENCE AND PUBLIC SECTOR
The e-Hearing backbone for the Ministry of Justice, the digital training infrastructure for the Ministry of Environment and an end-to-end transformation for the Ministry of Culture — critical national projects. The strategic projects run with HAVELSAN are the strongest evidence of our closed-network, on-premise AI capability.
AREA 02 · TELECOM AND AVIATION
The vast customer networks of globally operating brands such as Turkcell, Türk Telekom and Turkish Airlines are optimised with autonomous systems: speech-to-text, call centre performance analytics and intelligent interview and scoring models.
AREA 03 · INDUSTRY AND FINANCE
Industrial ERP architectures from industry giants such as ENKA down to SME factories; fintech solutions for financial institutions; and end-to-end marketplace platforms at the scale of Trendyol, from vendor management to payment systems and logistics — built from scratch.
Our HR Tech platform Hiri was honoured with the international Brilliance Award in London.
GET STARTED
You do not need to arrive at the discovery call prepared. Tell us which work is wearing you down, and we will get to a measurable problem statement together.
Book a discovery call →We would like to use Google Analytics to understand how the site is used. It sets cookies and transfers data to the United States. The site works exactly the same if you decline.
