Automation · Quality · Complaints · ERP
I connect the systems you already run with n8n and AI models, and take re-typing data, chasing deadlines and assembling reports off your team's plate. Eighteen years in manufacturing and quality mean I know which process is worth automating and which one needs fixing first.
Example flow log
Illustrative example on anonymised data.
Automation catalog
Twelve flows I build most often. Each starts with a concrete signal and ends with something you can check: a case in the register, a report, an alert. Pick an area or browse them all.
An email with attachments becomes a numbered case with a full data set and a confirmation to the customer, before anyone opens it.
The operator fills three fields on a tablet; the automation attaches the rest: photo, workstation, batch and the right person to notify.
No deadline passes quietly. The automation reminds the case owner and, once it is missed, escalates upwards with the full context.
A document from a supplier or customer is checked for required fields and values before it enters the workflow. Gaps go back to the sender immediately.
A 3.1 certificate or material cert is read, compared with the product requirements and saved against the batch in ERP.
A change to a controlled document triggers notifications and collects read confirmations. The auditor sees who, when and which version.
The team keeps working in Excel, but records are validated and reconciled with ERP both ways. No more moving data from a file into the system.
Items, suppliers and BOM structures are checked every night for gaps, duplicates and inconsistencies. The data owner gets a fix list in the morning.
On Monday morning management gets the week's cause distribution and the trend against the previous period. Nobody assembles it on Friday after hours.
Corrective actions do not close on a table entry alone. The automation reminds about the due date and, afterwards, asks for proof of effectiveness.
When a case moves to analysis, sections D1–D3 are filled from the register and D4 gets a draft from similar cases. The team starts from content, not from a blank template.
An engineer asks about a product, a defect or a customer and gets an answer with sources: similar cases, their causes and the actions that worked.
The catalog is not a price list or a set of off-the-shelf products. Every flow is built for a specific process and a specific client's data, and the substantive decisions stay with people.
Don't see your process? Describe itHow lasting improvement happens
A complaint, a process deviation or a customer escalation is only the input. Value appears when that signal turns into evidence, the real cause is found, and it is closed with an action whose effect can be measured. Automation does not replace that work; it takes the repetitive part off it.
Pick a step to see the artifact it produces.
Signal or problem. The case arrives with a full data set rather than as an email with no attachment. That decides whether analysis can start at all.
Illustrative example on anonymised data. The figures show a typical course, not a specific job.
Integrations
Automation does not start with a new system. It starts when ERP, email, spreadsheets and documents begin talking to each other, and people stop being the link between them.
Flows run on your infrastructure or in a cloud you choose. AI models receive only what a task needs, on terms written into a data processing agreement.
I do not replace your ERP or your mailbox. I plug into what is there and take moving data between windows off people's hands.
The automation prepares, classifies and reminds. The decision on a complaint, a cause or an action is made by the person accountable for it.
Solutions
Both came out of a concrete problem in a quality process, not out of an idea for a product. Both plug into the automations from the catalog.
Four things this screen enforces
Five steps instead of one text box
The reporter is walked through the form, so they do not get to decide what matters.
The invoice is mandatory
Without proof of purchase the case comes back anyway. Better to stop it here than a week later.
A summary before sending
Everything in one place for the reporter to check before the case starts moving.
Attachments counted and confirmed
The system shows what actually arrived, so nobody assumes the photos are somewhere on a server.
DN_supplier_0417.pdf
1 gap. Gap list sent to the sender, document on hold.
Portfolio
Products for companies, tools for my own work and websites for local businesses. Each one carries its state as of today: what is publicly up, what is available after signing in, and what is an internal tool or dormant.
Rollout management app for ERP Serwis, a Comarch ERP XL implementation company: at-risk projects, blockers, owner decisions, consultant workload and each person's week in one place.
Decision platform with three roles: an execution AI proposes a diagnosis and a fix, a validation AI challenges the assumptions, a human approves. The same model never proposes and approves at once.
Eight autonomous agent teams working on Solvio: deployment with rollback, QA regression, incidents, billing audit, security, database migrations, translation quality and social content. Run from a terminal, cron or n8n.
A self-hosted n8n instance with flows for Solvio and client websites: AI complaint triage, contact forms, notifications, reports. The same workshop where I build the flows from the catalog above.
Multi-brand pipeline for short PL and EN videos: AI scripts, schema validation, rendering in Creatomate, publishing across several platforms.
An MCP server connecting AI assistants to Google Sheets and Docs: reading and writing ranges, creating documents, working on team data without manual copying.
Website for a cleaning company from Częstochowa: services, price list, a contact form wired to an automation, and local SEO.
Website for a mortgage and business loan advisor in Warsaw: offer and consultation booking.
Website for a lash and brow studio, built as a demo for a local business.
Links go only where something is publicly up on the day of publishing. Projects for closed teams are shown behind a sign-in or described without a link; details and demos on request.
Expertise
I run each of them end to end: from assessing the current state, through delivery, to measuring effectiveness and stabilising the result.
Less work copied from one hand to another. The team makes decisions instead of moving data between files.
A system that mirrors the real process, not the one from the slide deck. People use it because they helped shape it.
An orderly complaint process cuts response time and takes the chaos off the team instead of adding one more spreadsheet to fill in.
A root cause closed for good instead of a recurring defect that comes back every quarter under a different name.
A quality management system that supports the work instead of blocking it, and holds up in a customer audit.
Process
I start by understanding the problem and the data behind it. The solution gets designed together with the people who will use it, because a standard only holds when it is theirs.
Talking to the team and management, mapping the constraints: people, systems, deadlines and budget.
Mapping how the process actually runs, not how the procedure describes it.
What gives the largest effect for the smallest cost, in what order, and who owns what.
Working with users, piloting, training, correcting based on what practice shows.
Checking whether the change actually worked, then locking it into the standard.
Track record
The figures below come from my work in manufacturing companies and from projects I ran. I do not publish metrics I cannot document.
18+years
in manufacturing, quality and projects
10,000+
customer complaints handled
over 2,000 a year across 5 years as Customer Quality Manager
~25%
less manual work in quality processes
after standardisation and automation were rolled out
~25%
faster response to customer cases
by putting the handling process in order
About
Quality management · ERP rollouts · Process automation
I spent 15 years inside manufacturing: quality, complaints, processes. I know the shop floor, the warehouse, and what a conversation with a customer sounds like when something has gone wrong. For five years I led a Customer Quality team handling over two thousand complaints a year from customers in more than a dozen countries.
For the past four years I have been running a global Infor M3 ERP rollout as Product Owner and Global Process Owner for quality. I plan schedules, coordinate consultants and business teams, run workshops, testing, data migration, go-live and stabilisation. I work in Scrum, in Azure DevOps.
Recent years have also been about building my own tools. I automate what can be automated and check where AI genuinely shortens the work rather than just adding a fashionable cost. Having all three perspectives at once, manufacturing, quality and technology, is rare.
Questions
No. Automation connects the systems you already have. If a system has an API, a file export or even just a mailbox, it can be plugged in. Replacing an ERP is a separate decision and a separate project, and it is not where automation starts.
Flows run on your infrastructure or in a cloud of your choice within the European Union. AI models receive only the fragments a task needs, and the scope is written into a data processing agreement. Complaint data is not used to train models.
For part of them, yes: reading documents, classifying cases, drafting reports, searching the case history. For deciding on the root cause and the action, no; that stays with the engineer. Before deploying anything I check on your data where AI gives a shortcut and where it is only a cost.
That is the most common starting point, and not a bad one. The first step is a map of how the work actually flows and picking one bottleneck, most often case intake. Excel stays as long as it is needed; the automation connects it to the rest.
Every flow gets documentation, an owner on the client side and error monitoring. I can keep maintaining it or hand it over to your IT team. I do not build solutions only their author understands.
A conversation about the process and the data that describes it. Then a short assessment of the current state and an action plan with priorities, as in the Process section. No commitment up front and no promising results that cannot be measured.
Contact
Write what is happening in the process, where data is re-typed by hand today and what you have already tried. I reply within one working day. If it is urgent, say so.
