# Rok Benko · Custom AI solutions for your business

**The engineer 3M+ developers learned AI from**

Fire your AI consultant. Hire an AI engineer.

I design, build, and run custom production AI solutions for European small and medium companies. Not tied to any AI company, and if you want, no data goes out of the company.

## Problems

You have had the AI meetings and the email inbox full of offers. Most of them are the same four things wearing different logos. Here is what they look like from the engineering side, which you don't see.

- **Meetings that go nowhere:** Call after call about what AI could do. Everyone nods, another meeting lands in the calendar, and the doing never starts. You pay for talk, not for a working solution.
  - In plain words: Some AI consultants make money from meetings, not from results. Every call is about what AI could do one day. Nothing gets built, because building isn't what they sell.
  - What AI consultants do: They book a monthly call to talk about your AI strategy. After six months you have notes, plans, and invoices, and your team still does everything by hand.
  - What I do: On our first call I pick one task that eats your team's time. The next thing you see isn't another plan. It is that task running for real.
  - Storytime: Your kitchen tap drips. One plumber comes every week, looks at the tap, talks about the tap, and sends a bill. The tap still drips. Another plumber comes once, fixes the tap, and the dripping stops. AI consultants are often the first plumber. I am the second one.
- **Workshops that change nothing:** You get a training day, an inspiring deck, and a certificate. Three months later nothing in your business runs differently. Knowledge fades, processes stay manual.
  - In plain words: A workshop is a school day for your team. People learn what AI is and go back to work. But listening about AI doesn't put AI into your company. Someone still has to build it, and after the workshop that someone is you.
  - What AI consultants do: They teach your team for a day, hand over the slides, and leave. Building the actual solution is now your team's homework, next to their normal jobs.
  - What I do: I don't leave you homework. I build the solution, connect it to your daily work, and run it for you. Your team just uses it.
  - Storytime: You want a warm dinner. A teacher comes, shows pictures of food, explains how cooking works, and leaves a recipe book on the table. At dinner time the kitchen is still cold. I don't leave recipe books. I cook the dinner and put it on the table.
- **Demos that never ship:** A chatbot that impressed everyone in the meeting and never touched your ERP, your email, or your real process. Demos are easy. Production is hard.
  - In plain words: A demo is an AI tool that works in a presentation, with clean examples picked in advance. Your real work is messier than that. Making AI survive real documents, real customers, and real exceptions is the hard part, and many projects never get that far.
  - What AI consultants do: They show a chatbot that answers three prepared questions perfectly. Connected to your real data it starts guessing, and the project quietly dies.
  - What I do: I build with your real documents and real processes from day one. The solution goes live in your company, and I stay to keep it running.
  - Storytime: You need a truck that carries heavy boxes through rain and mud every single day. A seller comes and shows you a shiny toy car, and it drives beautifully across the table. It looks great and carries nothing. A demo is the toy car. I build the truck.
- **API resellers taking a cut:** Someone else's AI model with a logo on top and a margin in the middle. When the AI provider changes prices or retires an AI model, that is suddenly your problem, not theirs.
  - In plain words: Big companies like OpenAI rent their AI out to anyone. A reseller rents it, puts a logo on top, and sells it to you at a higher price. The AI isn't theirs. The extra price is.
  - What AI consultants do: They stand between you and the AI company and take a cut of every bill. If prices change or an AI model is shut down, they shrug, because they didn't build anything.
  - What I do: I build the solution around whichever AI model fits you best, and I can change the AI model any time. You pay for engineering, not for a middleman.
  - Storytime: You want bread. A man comes, buys it from the baker for two euros, puts it in a bag with his own name on it, and sells it to you for four euros. If the baker raises prices or stops baking, that man has nothing to sell you. It was never his bread. I work differently. I cook the whole meal for you, and if one baker closes, I buy bread from the next one.

## What if AI actually worked for your business?

*Four building blocks, adapted per industry. On the page, visitors pick their industry and see four AI use cases written for businesses like theirs.*

### Agentic process automation *(Human approval built in)*
AI agents that run a real business process end to end: they read, decide, prepare, and hand the result to a person for approval. Weeks of manual teamwork become an automated run that costs a few dollars. I design the solution, build it, and run it for you.

### An assistant that knows your business
A knowledge assistant grounded in your documents, wikis, and internal systems, with answers that cite their sources. Your team stops interrupting the two people who know everything. Runs on EU infrastructure, isolated for your company.

### AI inside the ERP you already run
Invoice capture, order intake, and reporting you can ask questions in plain language, built around SAP, Microsoft Dynamics 365, or the local ERP you run today.

### Document pipelines that run themselves
Invoices, orders, contracts, and reports arrive as PDFs and email and leave as structured data in your systems. Judged on cycle time, error rate, and cost per document against your baseline, not on a demo.

### AI use cases by industry

- **Accounting & bookkeeping:** Invoices that book themselves · Ready for e-invoicing in 2028 · Costs coded to the right place · Management reports that draft themselves
- **Carpentry & custom furniture:** Cut lists with less waste · Renders that win approvals · Quotes from drawings, faster · Offcuts that get remembered
- **CNC & metalworking:** Quotes from CAD in minutes · Unprofitable parts flagged early · RFQs answered first · Past jobs price the next one
- **Marketing & creative agencies:** Client reports that write themselves · Ad variations at volume · First drafts against the tone guide · Your past work, searchable
- **Physiotherapy:** Notes written during the session · Home exercises that actually happen · A calendar that fills itself · Paperwork pre-checked
- **Medical clinics & private practices:** No-shows caught before they happen · Visit notes without the typing · Intake summarized before the consult · Recalls that reach the right patient
- **Law firms:** Clauses extracted across the whole stack · Deadlines that surface themselves · First drafts from your playbook · The firm's memory, searchable
- **Logistics, freight & customs:** Shipping documents that type themselves · HS codes suggested, not guessed · Where is my shipment, answered · Proof of delivery closed same day
- **Beauty & cosmetic salons:** Clients who come back on schedule · No-shows rescheduled in one tap · Bookings taken while you work · Win-backs that feel personal
- **Cleaning services:** Quotes from photos in minutes · Routes that respect the map · Quality checked from photos · Contracts that renew themselves
- **Auto repair shops:** Findings customers actually approve · Missed calls turned into bookings · Parts invoices reconciled to the job · Service reminders per vehicle
- **Restaurants:** Order what you will actually sell · Waste flagged where it happens · Menu decisions from data · Prep tuned to reservations
- **Hotels & accommodation:** Prices that move with demand · One-night gaps filled · Upsells before arrival · Reviews read for you
- **Hairdressers & barbershops:** Bookings taken mid-haircut · Cancellations filled from the waitlist · Rebooking on each client's cycle · Reminders that cut no-shows
- **Real estate agencies:** Listings written in seconds · Leads qualified while you sleep · Photos polished automatically · Seller packs assembled on demand
- **Opticians:** Recalls that know who is due for what · Empty exam slots refilled · Lens options matched to the person · Stock that orders itself
- **Architecture firms:** A second reviewer before submittal · Code questions answered with citations · The office's knowledge, searchable · Sheet indexes that check themselves
- **Driving schools:** Scheduling without double-booked cars · Hours tracked toward the test · Reminders that save lessons · Theory practice that adapts
- **Dental practices:** Claims that get paid the first time · Coverage checked without phone calls · Recalls and no-shows handled · Findings documented consistently
- **E-commerce & online shops:** Descriptions at catalog scale · Search and recommendations on-site · Returns explained by data · Visible to AI shopping assistants
- **Veterinary clinics:** Notes and invoices from the consult · After-hours calls captured · Inventory tuned to the season · Reminders per patient
- **Pharmacies:** Prescriptions without the retyping · Phone questions off the counter · Paperwork chased automatically · Stock that neither expires nor runs out
- **Fitness studios & gyms:** Churn caught while you can still act · Trials answered in minutes · A timetable shaped by demand · Win-backs that sound human
- **Insurance brokers & agencies:** Renewals compared line by line · Claims intake without re-keying · Coverage questions answered with sources · Renewal dates that never slip
- **Tax advisory & financial planning:** Research memos with citations, fast · Client documents into workpapers · Rule changes flagged per client · Conclusions in client language
- **Plumbing, HVAC & electricians:** The 11pm emergency call, answered · Paperwork from technician notes · Job to invoice without the gap · Jobs matched to skills and routes
- **Roofing & facade contractors:** Measurements without the ladder · A 3D model that sells the job · Storm damage triaged from photos · Material orders sized right
- **Construction & general contractors:** Takeoff in minutes, not evenings · Sub quotes leveled side by side · Bid documents drafted · Site progress from photos
- **Landscaping & garden services:** Quotes priced from satellite · Routes with less windshield time · Seasonal upsells on schedule · Weather-proof scheduling
- **Print shops & signage:** Bad files caught the moment they land · Complex jobs quoted instantly · Job tickets that route themselves · Proof cycles that keep moving
- **Event planners & venues:** Proposals while they are still excited · Enquiries qualified by fit · Vendor coordination on rails · Quick questions off your plate
- **Catering:** From signed quote to kitchen sheets · Quotes out while interest is hot · Dietary notes that cannot get lost · Follow-ups that rebook
- **Bakeries & artisan food:** Bake what tomorrow will sell · Purchasing tied to the plan · Wholesale demand predicted · Waste tracked to its cause
- **Travel agencies & tour operators:** Itineraries assembled from booking PDFs · Traveler questions, answered always · Offers matched to past trips · Documents checked before departure
- **Recruitment & staffing:** CVs parsed, candidates engaged · Dormant candidates resurfaced · Interview notes summarized · Built with the EU AI Act in mind
- **Translation & localization:** Human attention where it moves quality · Projects that set themselves up · Terminology enforced everywhere · An engine tuned to the domain
- **Funeral services:** Obituary drafts from case details · Hand-written sheets read automatically · Arrangement meetings captured · Tribute pages assembled with care
- **Pest control:** Recurring routes planned as one · Chemical logs kept compliant · High-intent calls never missed · Contracts flagged before they lapse
- **Moving & self-storage:** Surveys from the customer's own phone · Storage rates that follow demand · After-hours rentals captured · Move-day volume verified
- **Notaries & legal administration:** Deeds pre-filled and cross-checked · Checklists collected up front · Templates and clauses, searchable · Registry data checked against the file

## Architecture: When the AI world changes, my solution doesn't break.

Everything specific to your business lives in layers you keep. The AI model underneath is a component with a socket. When prices, quality, or rules change, the component changes, and your solution keeps running.

**Yours, stable:**

- **EU cloud:** The default: your solution runs on EU infrastructure I operate, isolated for your company. You buy no hardware and manage nothing, I run it all for you.
  - **Your workflows:** Approval chains, business rules, and the way your company actually works. Encoded once, stable across every AI model.
  - **Your integrations:** ERP, CRM, email, and databases, connected through documented interfaces. They do not care which AI model is underneath.
  - **Your data:** Grounding, retrieval, and history, isolated for your company and exportable at any time.
- **On-premise:** For companies whose data must not leave the building, the same solution runs on your own hardware as a premium engagement. The architecture stays the same, only the location changes.
  - **Your workflows:** Approval chains, business rules, and the way your company actually works. Encoded once, stable across every AI model.
  - **Your integrations:** ERP, CRM, email, and databases, connected through documented interfaces. They do not care which AI model is underneath.
  - **Your data:** Grounding, retrieval, and history, isolated for your company and exportable at any time.
- EU cloud vs. on-premise: Both options run the exact same solution. The only question is where the computers stand and who takes care of them.
  - EU cloud: Your solution runs on computers in a professional European data center, rented and operated by me. You buy nothing and maintain nothing, it works from day one and I watch it around the clock. You pay a monthly running cost. Your data does leave your building, it lives on those EU computers, isolated for your company, never mixed with anyone else's and never used to train AI models. For most companies this is the comfortable default.
  - On-premise: On-premise means on your premises: the solution runs on computers standing inside your building. You buy the hardware once and pay for its electricity, and I set everything up and keep running it for you. Your data never leaves the house, which is why companies with strict rules or business secrets choose this. It costs more up front and starts a little slower, and in exchange nothing about your business ever travels anywhere.
  - Storytime: Your house runs on electricity. Now let's say all that power had to live somewhere, in a battery, with all the wiring around it. The EU cloud is like keeping that battery in my guarded hall: I built the shelves, I watch the meters day and night, and you pay a small monthly rent for the space. On-premise is like hanging the battery in your own cellar: you buy it once, and everything your house makes stays inside your walls. Same battery, same power. The only question is whose walls stand around it.
- **The frontier AI model, swappable:** OpenAI, Anthropic, Gemini, Grok, Mistral. Paid per use. Your data goes out of the company.
- **The open-source AI model, swappable:** DeepSeek, Kimi, GLM. Free to use, runs on dedicated hardware. Your data never reaches an AI provider.
- OpenAI today, Anthropic tomorrow, Mistral or an open-source AI model after that. The swap is my job, and it costs you nothing extra.
- In plain words: Most AI tools are built directly on one company's AI model. If that company raises prices or shuts the AI model down, the tool breaks. I build your solution in layers: your business rules and connections live in layers that stay, and the AI model sits in a socket underneath. We can unplug one AI model and plug in another without rebuilding anything.
- What AI consultants do: They hard-wire your solution to one AI model. When the price doubles or the AI model retires, you pay again or start over.
- What I do: I keep the AI model in a socket. When a better or cheaper AI model appears, I swap it as part of the service, and your solution keeps running.
- Storytime: Your light bulb burns out. You don't call anyone to rewire the house. You unscrew the old bulb, screw in a new one, and the light is back. The AI model is the bulb. Your solution is the wiring, and I built it so every bulb fits.
- Frontier vs. open-source AI models: Every AI tool has an AI model underneath. It is the engine that reads, thinks, and writes. There are two families of AI models, and the choice between them decides what you pay, how fast it runs, and where your data goes.
  - Frontier AI models: Frontier AI models are the strongest ones, built by big companies like OpenAI, Anthropic, and Google. ChatGPT and Claude run on frontier AI models, so if you have used those, you have used a frontier AI model. Nothing gets installed at your company: your question travels to the AI company's computers and the answer travels back. That is the catch, your data goes out of the company. You pay for exactly what you use, measured in tokens, which are small pieces of words, roughly three quarters of a word each. A few cents per task sounds cheap, but heavy daily use adds up every month. In return you get the smartest and usually fastest AI for hard tasks. The dependence is real: the AI company sets the prices and the rules, and it can retire an AI model you built on.
  - Open-source AI models: Open-source AI models like DeepSeek, Kimi, and GLM are published for anyone to download and run. The AI model itself is free. What costs money is the hardware, a powerful computer with a strong graphics card, and the work of setting it up and keeping it running. Because the AI model runs on your machines, your data never goes out of the company. Speed depends on your hardware: a strong machine answers quickly, a weak one crawls. For most everyday business tasks the best open-source AI models sit close behind the frontier ones. And nobody can raise the price or shut down an AI model you own a copy of.
  - Storytime: You already know the story: your house needs electricity every day. Now the question is where the power comes from. A frontier AI model is like power from the big electricity company: always strong, always on, but a bill arrives every month, and the company sees exactly how much you use. An open-source AI model is like solar panels on your own roof: the sunlight is free, you only pay once for the panels and someone to install them, and then the power is yours. Nobody sends bills, and nobody sees what you use it for. On dark winter days the power company is stronger. But most days, your own roof does the job. I know how to connect both, and I choose what fits your house.

## Comparison

### Chatbots in the browser vs. custom solutions I design, build, and run for you

ChatGPT, Claude, Gemini, Copilot, and Grok are excellent personal assistants. Keep them. This is what they cannot be.

| | ChatGPT / Claude / Gemini / Copilot / Grok | My custom solutions |
| --- | --- | --- |
| Purpose | General assistant for whoever is typing that knows a bit of everything but masters nothing | Software built around your exact workflows that does your specific work on its own, day and night |
| Privacy | Everything you paste goes to the provider's cloud, under terms they can change at any time | EU cloud isolated for your company, or your own hardware, and your data is never used for training |
| Integrations | Your employees copy data in and paste results out by hand, all day, with mistakes | Connected directly to your ERP, CRM, email, and databases, so nobody retypes anything |
| Automation | Employee still writes prompts, reads answers, and rechecks everything, so the work still needs a person all day | Agents run the whole process end to end, your people only approve the results |
| Consistency | Answers change depending on who asks and how, so quality is a lottery | Same tested process every time, with logs you can check |
| AI model changes | Provider changes or retires an AI model and your workflows break overnight | Swap is my job and costs you nothing extra, your work never stops |
| Accountability | Help center that might not get back to you in a few days, if at all | One engineer who built it, runs it, and picks up the phone when something feels off |
| Pricing | Per seat, per month, forever, and the bill grows with every employee you hire | One agreed project price, then a flat service fee per solution, no matter how many people use it |

### Working with AI consultants vs. working with me

Most AI offers you have received differ in logo only. Here is where this one differs in substance.

| | AI consultants | Me |
| --- | --- | --- |
| Outcome | Slides, a workshop, or a demo, and after the invoice nothing in your company runs differently | Working software in production, built for your process and then run for you |
| Durability | Knowledge that fades in weeks and a demo that quietly dies | Solution that keeps running for years, operated by the person who built it |
| Dependence | Everything built on the one API they resell, so their vendor's problems become your problems | Independent of any single AI company, swaps included in the service |
| Privacy | Your data goes wherever their tool stack requires, often outside the EU | EU jurisdiction, isolated for your company, exportable any time you ask |
| Adaptability | Market moves and you get a new offer for a paid rebuild | Market moves and I absorb the change as part of the service |
| Exit | Leave and you start over from zero, keeping nothing | Your data and integration documentation leave with you, and a source-code buyout option exists |
| Results | Success means documents delivered, not problems solved | Measured in your numbers: cycle time, error rate, and cost against your baseline |

## Steps

1. **Scope:** We talk about your business: what you do, what you've already tried, and where manual work eats the most time. Together we pick the pain points worth solving and check what's possible with your data.
2. **Prototype one workflow:** This is where I build your custom AI solution. We choose one high-value workflow and I build it from start to finish, with human approval on anything that writes to your systems, so you stay in control.
3. **Measure AI output:** We look at what the prototype produced: the files, the answers, the finished work. Then we judge how close we are to automating the manual work and how much time and money that would save you.
4. **Run & extend:** I run the solution for you: hosting, monitoring, AI model swaps, and extensions as your needs grow. Your data stays exportable at any time and a source-code buyout option exists.

## Experience

### Weeks → 3 hours: Agentic automation
*Stealth Startup, AI engineer*

Built an AI-powered MVP that turned weeks of manual teamwork into a fully automated 3-hour process. Cost: just $20 per run.

### 4 hours → 3 minutes: Agentic automation
*PodcastPrepper, founder*

Built PodcastPrepper, an AI agent that researches a podcast guest, creates interview talking points, and delivers a full prep report in 3 minutes. Cost: $0.13 per run.

[See the open source code](https://github.com/rokbenko/ai-playground/tree/main/crewai-tutorials/1-Podcast_prepper/podcast_prepper)

### #1 for OpenAI: Community authority
*OpenAI community, top contributor*

Became the #1 OpenAI API contributor of all time on Stack Overflow with 23,800+ reputation points and 519 answers. The go-to person when millions of developers were first learning to build with AI.

[Verify I'm not lying](https://stackoverflow.com/users/10347145/rok-benko?tab=topactivity)

### Top 1% for CrewAI: Community authority
*CrewAI community, top contributor*

Ranked in the top 1% on the CrewAI contributor leaderboard with 281 answers and 22 accepted solutions. Also got a pull request merged into the official CrewAI docs.

[Verify I'm not lying](https://community.crewai.com/u/rokbenko/summary)

## Story

I got into AI in December 2022. Not because it was trending. Because I made my first OpenAI API call and couldn’t sleep that night.
Back then, nobody was calling themselves an AI expert. There were no LinkedIn posts about “5 AI tools that will 10x your business.” There were just a few developers, a half-broken API, and a lot of unanswered Stack Overflow questions. I started answering them. Every single day. By the time AI got popular, I was already the #1 OpenAI API contributor on Stack Overflow with 3M+ people educated.
I’m not telling you this to brag. I’m telling you this because it matters who builds and runs the solutions your business depends on.
Right now, the AI space is full of AI consultants who found ChatGPT one month ago. They run meetings that go nowhere. They charge $5,000 to teach your team “prompt engineering.” They resell someone else’s API with a logo on top. They’ve never fixed a broken AI agent at 2am. Never had to redo a whole setup because a provider raised prices overnight by 10x.
I have. Every day, since 2022.
I know which tools promise too much and don’t deliver. I know which frameworks looked good and quietly died. I know how OpenAI’s Assistants API went through two betas before being replaced. I know how price changes have made entire solutions outdated overnight. That is exactly why every solution I build is independent of the AI model: the AI model is a part you swap, not a foundation you marry.
And I don’t hand you a demo and disappear. I design the solution, build it, and then run it: hosting on EU infrastructure, monitoring, AI model swaps as the market moves, and extensions as your needs grow. If you want, it runs on your own servers instead, and no data ever goes out of the company. The solution works and keeps working, without you hiring an AI team.
What I offer isn’t theory. It’s hard-won engineering from someone who was here before AI was cool, and stayed because I genuinely love the work.
If you want hype, there are a thousand AI gurus who’ll happily sell it to you.
If you want someone who’ll tell you exactly what AI can and can’t do for your business, and then actually build and run it, let’s talk.

## Reviews

> Rok gave me the foundation to understand AI systems by breaking complex ideas into simple parts. It shifted the way I think about building at exactly the time I needed it most. The way he explains things makes them easy to understand, to the point that anyone can follow. He understands the fundamentals so well that he can pass that knowledge to others. Being exposed to this early made all the difference for me, and I’ll always be grateful for it.
>
> Gustavo Gonzalez, Software engineer

*More reviews render on the page as screenshots.*

## Contact

Whether you need one process automated or a full solution designed, built, and run for you, let’s talk.

- [Book a free discovery call](https://calendly.com/rokbenko)
- Email: [roksstartups@gmail.com](mailto:roksstartups@gmail.com)

## FAQ

### What kinds of solutions do you build?

Four kinds: agentic process automation, knowledge assistants grounded in your data, AI modules for the ERP you already run, and document pipelines. Each is built around your workflows, connected to your systems, and then run for you as a service. Pick your industry in the Solutions section to see what they look like for businesses like yours.

### What makes you different from AI consultants?

An AI consultant sells you advice: meetings, workshops, maybe a demo that never ships. Then they leave, and running anything is your problem. I do the opposite. I design the solution, build it, and then run it for you: hosting, monitoring, AI model swaps, and extensions as your needs grow. The comparison above puts the two side by side.

### What happens to our data?

It stays yours. The solution runs on EU infrastructure, isolated for your company, and your data is never used to train AI models and never mixed with anyone else's. You can export it at any time. For data that must not leave the building, the solution can run on-premise, on your own servers.

### We already use ChatGPT and pay for Copilot. Why this?

Keep both. They are personal productivity for whoever is typing. This is company infrastructure: software connected to your ERP, CRM, email, and databases that runs processes end to end with approvals and logs, and reaches the customizations generic copilots cannot.

### Aren't we just locked into you instead of OpenAI?

Honestly: you depend on me the way you depend on any software vendor. Three differences. The exit is defined up front, with your data and integration documentation always exportable and a source-code buyout option. The solution is never hostage to one AI company's roadmap, AI model swaps are included. And you are dealing directly with the engineer, not a reseller of someone else's platform.

### You are one person. What if something happens to you?

A fair question for any vendor. Operations are documented and monitored, the stack is widely used open source rather than private code only I know, and your data and integration documentation stay exportable at any time. The quiet truth: big vendors sunset products too, usually with less notice.

### Which AI models do you use?

Whichever fits the task. Frontier AI models like OpenAI, Anthropic, or Gemini when you need the strongest quality. Open-source AI models like DeepSeek, Kimi, or GLM when data must not leave the house or costs must stay flat. Your solution is not tied to any of them, and I swap AI models as the market moves.

### How does an AI model swap actually work?

The solution is built against a layer independent of the AI model, so a swap is a contained engineering change: re-test, re-tune, roll out. I do it as part of the service when prices, quality, or regulation make a better option available, whether that is another frontier AI model or an open-source one. The swap is my job, not your project, and it costs you nothing extra.

### What if the AI makes a mistake?

Honestly, sometimes it will. That is why anything that writes to your systems goes through human approval: the AI prepares the work, your people keep the final word. Every action is logged, and I watch the solution in production as part of the service.

### What does working with you look like, and what does it cost?

One agreed project price for a prototype on one workflow. We measure what it produces and how much manual work it takes off your team. Then a flat service fee per solution, covering hosting, monitoring, AI model usage, and extensions, no matter how many people use it. Running costs are stated up front, and nothing is priced per seat.

### Who owns the code and IP?

I do. That is what makes the price an agreed project plus a flat service fee instead of an enterprise software license. You own your data, your configurations, and your results, all exportable. And if you ever want the solution fully in-house, a source-code buyout option exists.

### Is this GDPR compliant?

The architecture is designed to support compliance: EU jurisdiction, audit logging, and human approval on anything that writes to your systems. Compliance itself is a shared responsibility that depends on how the solution is operated. I build with that in mind rather than promising a rubber stamp.

### Who is this for, and who is it not for?

European small and medium companies with real processes and real data, typically without an internal AI team, served in English, Slovenian, and German. It is not for companies that want a workshop, a rebranded ChatGPT, or AI writing to financial systems without human approval.

### Who operates this business?

KAPION d.o.o., registered in Slovenia, VAT SI32880227. Rok Benko is a co-owner and does the work himself.