IT Solutions — Dawid Zmarzlak
Home/AI & automation

AI implementation and process automation for businesses

I build AI assistants, automations and language-model integrations where they actually save your team time. I start with an audit of the process, then a pilot on one task, and only then the full rollout. Where n8n or Make is enough, I don't write code. Where code is needed, I write it properly.

Book a free 30-minute call →See prices ↓

Net prices, visible from the first contact. The audit fee counts towards the implementation.

// four lines

From audit to assistant.

Each line has a starting price and a typical range. I give the exact figure after a call, because it depends on how many systems and how much data need to be connected.

[01]

AI audit and consulting

You know which processes are worth automating, what it costs and what pays back.

I go through your company's processes with you, find where time leaks, and assess what can safely be handed to automation or AI. You get a concrete list with priorities, estimated costs and risks. The audit fee counts towards the implementation.

Scope
  • Workshop with the people who run the process every day
  • Process map with a verdict: automate, support with AI, or leave as is
  • Estimated build and running costs
  • Risk review: GDPR, the AI Act, data quality
  • Tool recommendation: n8n, Make or custom code
Tools
  • Team workshop
  • Process map
  • Cost estimate
From
1,900 PLN net
~440 EUR
typically 1,900-3,500 PLN net
Time
1-2 weeks
[02]

Process automation

Repetitive work runs on its own, and your team focuses on what can't be automated.

Workflows that move data between systems, send notifications, generate documents and keep track of deadlines. Where the content has to be understood (an email, an invoice, a ticket), I add a language model. The price is per process.

Scope
  • Analysis and a step-by-step breakdown of the process
  • A workflow in n8n or Make, or a custom service when the tool isn't enough
  • Error handling and an alert when something goes wrong
  • Logs and a how-to for your team
Tools
  • n8n
  • Make
  • Gemini / Claude API
  • Webhooks / REST API
From
2,500 PLN net
~580 EUR
typically 2,500-8,000 PLN net per process
[03]

AI integrations in applications

Your application or system gains AI features: scoring, classification, summaries and draft replies.

I build language models into existing software: scoring and categorising tickets, extracting data from documents, preparing draft replies. With retries, a fallback provider and cost limits, the same way my own systems work.

Scope
  • Integration with a model API: Gemini, Claude, OpenAI or a local model
  • Fallback provider and retries when something fails
  • Daily limits and API cost control
  • Tests on real cases from your own data
Tools
  • Gemini / Claude API
  • Spring Boot
  • Next.js
  • PostgreSQL
From
5,900 PLN net
~1,370 EUR
typically 5,900-20,000 PLN net
[04]

AI assistant on your company data

Customers and staff get answers from your own documents at any hour, not ones the model made up.

A website chatbot or an internal knowledge base built on your documents (RAG): offer, price list, procedures, terms. It answers from the sources, introduces itself as AI, and when a question goes beyond what it knows, it saves the contact or hands the conversation to a person.

Scope
  • Preparing and indexing your documents (RAG)
  • A website widget or an internal assistant for your team
  • AI disclosure in line with Article 50 of the AI Act
  • Hand-off to a person and lead capture
  • Admin panel with conversation history
Tools
  • Gemini / Claude API
  • RAG (pgvector)
  • Spring Boot / Next.js
  • n8n
From
7,900 PLN net
~1,840 EUR
+ care from 390 PLN / mo
typically 7,900-15,000 PLN net + care plan
// running in my stack

Already in production in my business.

Before I offer something to a client, I use it myself. Three systems that run in my business every day. No made-up numbers: just what they do and what they are built with.

The chat on this website

Problem

Visitors ask about prices and timelines at any hour, and I don't reply at night or in the middle of client work.

Solution

The AI assistant in the corner of the page answers questions about my services, prices and timelines, introduces itself as AI and points people to the right page. When someone wants a quote, it collects their contact details and passes them to my admin panel. The mechanism works to set guidelines: it sticks to my offer and stays on topic.

Stack
  • Language model (LLM)
  • Backend API
  • Admin panel

Job aggregator with AI scoring

Problem

Going through several job and freelance portals every day took time, and most of the listings didn't fit me at all.

Solution

The system regularly collects listings from several portals, merges duplicates and scores which ones fit my profile. For the chosen ones it prepares a draft reply. The mechanism works to set guidelines, and the decision and sending stay with me.

Stack
  • Language model (LLM)
  • Portal data collection
  • Backend API

Follow-up sequence after an enquiry

Problem

Some people ask for a quote and go quiet. Following up with each of them by hand a few days later is easy to forget.

Solution

After an enquiry the system sends follow-ups by itself a few days apart, only during business hours and with one-click unsubscribe. The mechanism works to set guidelines. This is plain automation with no AI: not every process needs a language model.

Stack
  • Job scheduler
  • Email
  • Backend API
// hybrid

When n8n, when code.

I don't sell one tool. I pick it to fit the scale, the systems you already have, and what upkeep should cost.

When an off-the-shelf tool is enough

n8n or Make

  • Moderate volume: tens or hundreds of runs a day.
  • Systems with ready-made integrations: email, spreadsheets, CRM, messengers, forms.
  • The process changes often and you want to adjust it without a developer.
  • Lowest starting cost: I set up n8n on your server, with no per-run fees.
When no-code stops being enough

Custom code

  • High volume or strict response-time requirements.
  • Systems without ready connectors: an older ERP, KSeF e-invoicing, your own databases.
  • An assistant on your documents (RAG) with control over what the model sees and cites.
  • Cost measured in years: at scale, custom code is often cheaper than subscriptions.

I often combine both: n8n wires up the simple steps, and a custom service handles what n8n can't.

// process

Audit, pilot, rollout, care.

  1. 01

    Audit

    I talk to the people who run the process, check how much time it takes, and point out what is worth automating and what isn't. You get a report with priorities and a cost estimate. The audit fee counts towards the implementation.

  2. 02

    Pilot

    One process or one assistant on a limited set of data. We test it on real cases before you invest more.

  3. 03

    Rollout

    The full version: integrations, permissions, logs, tests and a how-to for your team. Accounts and code are in your name from day one.

  4. 04

    Care

    Monitoring, prompt and rule tweaks, model and integration updates. We agree the scope of care at rollout, with no multi-year contracts.

// trust

No black box.

GDPR and a data processing agreement

When the solution processes personal data, we sign a data processing agreement (Article 28 GDPR). I choose model providers that offer data processing terms and tell you plainly where your data goes.

AI Act Article 50 as standard

Since 2 August 2026, a person talking to an AI system has to know it. Every assistant I deploy says so from the first message, and I label AI-generated content wherever the law requires it.

No lock-in

Provider accounts, the server, the repository and the API keys are in your name. You get documentation, so you can switch to another developer without starting from scratch.

Your choice of model

Gemini, Claude, GPT or a model running locally on your server. When data must not leave your company or the EU, I pick the model and hosting to match.

Transparent running costs

Before we start, I show you an estimate of the monthly costs: the server (usually a few dozen zloty) and the model API (from a few to a few hundred zloty, depending on traffic). You pay the providers directly.

A human has the final say

Where a mistake is costly, AI prepares a draft and a person makes the call. That is how my own job aggregator works: AI drafts the reply, I send it.

// FAQ

Questions about AI in your business.

How much does an AI implementation cost?

It depends on the scope. An audit starts at 1,900 PLN net, a single automation at 2,500 PLN per process, an AI integration in an existing application at 5,900 PLN, and an AI assistant on your company data at 7,900 PLN plus care from 390 PLN a month. On top of that come server and model API costs, which you pay the providers directly.

Do I need big data or an IT department?

No. Most small-business rollouts start with documents you already have (offer, price list, procedures, FAQ) and tools you already use. On your side I need one person who knows the process and can answer questions during the audit.

n8n, Make or custom code?

It depends on volume, systems and running costs. Simple flows between popular tools I build in n8n or Make. I write custom code for high traffic, for older systems without connectors (ERP, KSeF) or when an assistant has to work on your documents. I often combine both approaches.

What about GDPR and customer data?

We sign a data processing agreement, I choose model providers that offer data processing terms, and I keep what goes to the model to the necessary minimum. When data must not leave your company, I use a locally hosted model.

Does a website chatbot have to say it is AI?

Yes. Since 2 August 2026, Article 50 of the AI Act requires that a person talking to an AI system knows it, unless that is obvious from the context. Every assistant I deploy introduces itself as AI from the first message.

How long does a rollout take?

An audit usually takes 1-2 weeks, a single automation 1-3 weeks, an AI assistant 2-4 weeks, and an AI integration in an existing application 3-6 weeks. You get the schedule together with the quote.

First step: a 30-minute call.

Describe in a few sentences the process that takes your team the most time. I reply within 24 hours and suggest a time for a call. No obligation.

Book a free call →