Skip to content
Verhex
AboutServicesProjectsNotesContact
VERHEX

AI · Enterprise integration · Software

Precision software. Engineered solutions.

Links
  • About
  • Services
  • Projects
  • Notes
  • FAQ
  • Contact

© 2026 Verhex · All rights reserved.

hello@verhex.io
Straight answers

FAQ.

The questions we get asked most about AI and integration projects. Answers are short and direct.

About Verhex

What is Verhex and what does it do?
Verhex is an İstanbul-based AI and software engineering studio. It builds generative AI solutions (LLM, RAG, agent architectures) and integrates them into existing enterprise ERP, CRM, and data infrastructure. There are three service lines: AI, integration, and technical consulting.
Where is Verhex located?
İstanbul, Türkiye. We work remotely with clients across Türkiye and Europe, and meet in person within İstanbul.
How large is the team?
Small and senior. We staff projects with focused teams of 2–4 people. That is deliberate: in AI and integration work, a larger team does not add speed, it adds coordination cost.
Which industries do you work in?
We are not industry-restricted; the shape of the problem matters more. We add the most value with organizations holding large document archives, companies moving data between multiple systems, and software firms adding AI features to their product.

How we work

How do we start?
With a short intro call. We listen to the problem and the current state, then propose a feasibility study. Feasibility usually takes one week and produces an honest answer: can this be done, with which approach, and roughly how long it takes.
Do you work fixed-price or hourly?
Fixed fee where scope is clear — feasibility, consulting, defined integrations. For AI projects where scope is discovered along the way, we work on a monthly team basis. We tell you which model fits in the first conversation.
Who owns the code?
You do. All code, infrastructure definitions, and documentation belong to you and are handed over at the end of the project. We do not tie you to a closed platform of ours; when we step away, the system keeps running.
Do you provide support after the project ends?
Yes, in two forms: a defined maintenance agreement, or on-call support as needed. With AI systems the first three months matter most — as real usage data arrives, accuracy and cost need tuning.
Can you work alongside our existing team?
Yes, and it is the model we prefer. We work next to your team and transfer knowledge as we go. The goal is that by the end your team can develop the system without us.

Cost and timeline

What does a project cost?
It depends on scope. For reference: a feasibility study is one week of work, a working AI pilot 4–6 weeks, and an end-to-end enterprise integration 6–8 weeks. We give a concrete range after the first conversation — we do not send vague proposals.
What are the running costs of an AI system?
Two lines: model usage (per token) and infrastructure. During feasibility we produce a concrete monthly estimate based on expected volume. Cost-reduction levers — caching, small-model routing, prompt optimization — are built into the architecture from the start.
How soon can you start?
Usually within 2–3 weeks. Feasibility studies can start sooner since they need a small team and a short calendar.

Technology and security

Is our data safe? How is data protection handled?
We design the architecture around your data sensitivity. For personal data we use three approaches: masking before anything reaches the model, hosting in the EU region, or running the model entirely on your own infrastructure. Which data goes where is put in writing at the start of the project.
Which technologies do you use?
On the AI side we work with the major large language models and open-source models through a provider-agnostic layer. On the application side, mostly TypeScript, Next.js, Python, PostgreSQL, and container-based infrastructure. We pick technology by what your team can sustain, not by what is fashionable.
Cloud or on-premise?
Both are possible. We decide based on data sensitivity, your existing infrastructure, and cost balance. An open model on your own servers is a realistic option for organizations where data must not leave — but hardware and operating cost need to be discussed up front.
What happens when the AI answers incorrectly?
We build assuming it will. Three layers: showing the source of every answer so users can verify it, handing off to a human below a confidence threshold, and measuring accuracy with regular evaluation tests. The error rate never reaches zero; it becomes manageable and measurable.

Let's start

A short intro call is enough. We'll scope it together from there.

Get in touch