AI integration
Practical AI in your products and workflows: assistants, automation, less busywork.
- Typical stack
- Claude API, n8n
You get
I build AI into the products and workflows you already run: assistants that answer real questions, automations that handle the repetitive parts, and pipelines that read and sort documents so a person doesn’t have to. This is for teams drowning in the same inbox messages, the same lookups, the same copy-paste, who want the busywork gone rather than a demo that impresses for a week.
- Assistants grounded in your content
A chat assistant that answers from your own documentation and data. It takes the questions your team fields over and over, in the languages your customers actually write in.
- Workflow automation
The repetitive steps between your tools wired together with n8n and APIs, so routing, replies, and updates happen without anyone clicking through them.
- Document and data processing
Invoices, forms, emails and messy spreadsheets read, extracted and sorted into something structured your systems can use.
- Honest scoping first
I tell you up front which parts of the job AI actually improves and which are better left as plain software or a human, so you don’t pay for AI that’s only there to look modern.
- Guardrails and clear data boundaries
Rate limits, abuse safeguards, and a defined path for where your data goes and what the model is allowed to see. It’s all decided before anything ships, not bolted on after.
- Answers from your sources
Assistants are designed to answer only from your own sources and to say when they don’t know, rather than guess.
The win is quiet: the same handful of repeated questions and manual lookups stop landing on a person’s desk, so your team spends its hours on the work that actually needs a human. Where AI wouldn’t earn that, I tell you before you spend on it.
Your data
- Always introduced as AI
An assistant tells people they are talking to an AI, as the EU AI Act requires (Article 50).
- A processing agreement
A data processing agreement (DPA) with you and with the provider of the model. Data stays in the EU where the service allows it.
- Not used for training
I use model APIs whose terms say your data isn’t used to train their models.
No model is right every time, so each flow has a clear point where a person checks the result or takes over.
What it can look like
No public case study for this yet. Bring your use case to the call.
What to bring to the call
- The task that keeps coming back: what comes in, what a person does with it and how often.
- Two or three real examples, anonymised if you like: emails, documents or spreadsheets.
- Where the result should end up: which tool or system.
Who’s behind Blik
Jan Štok
Blik is a studio of one: I scope, design and build every project myself, so the person on your first call is the person who ships it. I’ve built hotel websites, a quoting system that replaced a distributor’s Excel and an iOS app for health data, and connected booking and smart locks for an overloaded reception. Whatever I build, you get a written handover, and on new projects the domain, hosting, repository and accounts are in your name from day one.
Three characters, one person. Blink listens, Scribble builds, Spark launches, and all three are me.
How it goes
Discovery & proposal
We map the real problem, constraints and what success looks like, then put the scope, fixed price and timeline in writing.
You get: a clear brief and a one-page proposal
30-min call · proposal within two working days
Design & build
You see working software early and often. Brand, interface, and engineering in one pair of hands.
You get: a working build
Sites 3–6 weeks, apps and platforms 6–12
Launch & support
Handover, training, and a single person to call when something needs attention.
You get: the running assistant or automation and a written handover
Questions
Is AI always the right answer?+
No, and I’ll tell you when it isn’t. AI earns its place only when it removes real, repeated work. If a simple form or a bit of plain software does the job better, that’s what I’ll build.
Where does my data go, and will it make things up?+
Only where we agree in writing: a processing agreement with you and with the model provider, data in the EU where the service allows it, and model APIs whose terms rule out training on your data. Assistants are designed to answer only from your own sources and to say when they don’t know. No model is right every time, so each flow has a clear point where a person checks the result or takes over.
Have you done an AI project I can look at?+
No public case study for this yet. The sample flow on this page shows the kind of work, not a client project; bring your own use case to the call.
How do we start?+
A short call or an email describing what you’re building. If we’re a fit, you’ll get a one-page proposal with scope, a fixed price and a timeline within two working days.
What does it cost?+
A fixed price per project, agreed before anything starts. No surprise invoices. Tell me the scope and I’ll send you a proposal.
You’re one person, so what if you’re unavailable?+
Fair question. The answer lies in how the project is set up, not in a promise. On new projects, the domains, hosting and repositories are in your name from day one, and the code and documentation are yours once the final payment is made. I build on mainstream tools, not a private framework, so another competent developer can pick it up. For anything in production we agree a response window in writing before it goes live.
