LAW / OPERATIONS / AUTOMATION / SOFTWARE

Understand the work first.
Build second.

I am most interested in work that is still done by hand simply because everyone got used to doing it that way. I look at how the work actually happens. Then I look for a better route: a workflow change, automation, a custom tool, or a few small parts that need to work together.

↓ What I can do Selected projects →

01 / WHAT I DO

What I can do for you.

I do not have a fixed menu of services. Bring me a concrete problem and I will work out what kind of intervention actually makes sense.

Some work is too specific for off-the-shelf software and too small to justify traditional custom development. That used to mean spreadsheets, email or manual workarounds. It does not have to stay manual anymore.

The solution does not have to become an app. It might be a script, an automation, a local utility, a small internal tool or a few connected parts. The job decides the format, not the other way around.

I can automate a manual process, connect applications, extract and structure information from documents, or build a tool exactly where generic software stops fitting.

AUTOMATION / INTERNAL TOOLS / DOCUMENTS / DATA / LOCAL APPS / INTEGRATION / WEB

I start with the problem, not the technology.

02 / PROJECTS

Selected projects.

W001
FLAGSHIP / ACTIVE

Neuroprotect Scribe

NEUROSONOLOGY / WINDOWS / LOCAL-FIRST / STT / OCR / MINDRAY ELITE I9

From voice and examination data to a cleaner factual backbone for the report. Less retyping, more time for the work that still belongs to the physician.

Problem

A specialist report starts with voice, but some of its facts already exist before dictation: in measurements, parameters and other outputs from the examination. Yet information still gets retyped, dictated again, corrected and assembled into the final documentation.

Tool

After several days observing the real workflow in a neurosonology laboratory, I built a local Windows tool for specialist dictation and documentation. It combines Czech speech-to-text, custom terminology, OCR and the surrounding workflow.

03 / LOCAL-FIRST

Sensitive data

With health data, privacy is not something added at the end. It is a design constraint.

Speech-to-text runs locally and sensitive data stays on the device instead of being sent through extra services simply because that would be easier for the technology. The aim is to keep processing close to where the data originates and preserve control over what happens to it.

What I am building next

The next layer is built around the Mindray Elite i9 ultrasound system. I am working out how much useful measurement and parameter data can be extracted reliably from its outputs, normalised into structured facts, and used to prepare the factual backbone of a medical report.

I do not want software to think for the physician. I want it to stop asking the physician to repeat work the machine has already done.

The physician remains responsible for review, interpretation and the conclusion. The software handles the transfer and organisation of facts that already emerged during the examination.

A living project

Neuroprotect is not a closed product. It is used in day-to-day work, and each new layer comes from what real use reveals.

Every extension has the same filter: remove another unnecessary step without losing control, traceability or privacy.

IN USE

Flagship project. The clearest example of how I work: observe the real operation first, build the tool second, then keep changing it as reality exposes the next useful improvement.

W002

Fire-Safety Documentation

DOCUMENTS / RULES / PROVENANCE / VALIDATION

Problem. Fire-safety documentation is assembled from multiple project sources. Information may repeat, conflict or be hidden in unstructured documents.

What I built. The workflow converts inputs into individual machine-readable facts, keeps their provenance, detects conflicts and separates interpretation from deterministic rules and calculations. AI helps where something has to be read and understood. It does not decide where a rule should decide.

Use. Used in real work by a professional authorised by the Czech Chamber of Authorized Engineers and Technicians Active in Construction; expert review remains part of the workflow.

W003

Digital Submission Converter

WINDOWS / BATCH / DOCUMENTS / IN USE

Problem. Digital construction submissions require electronic files in specific formats and parameters. With larger batches, manual preparation becomes repetition.

What I built. A simple Windows drag-and-drop utility converts documents in batches, preserves the originals, and checks structure, metadata, page count and dimensions. No platform. No account. No dashboard nobody asked for.

Use. Used in day-to-day work by an authorised professional.

W004

Visual Pipeline

IMAGE / SVG / VECTOR / HEADLESS / PIPELINE

Problem. My own visual workflow accumulated too many manual steps between different applications. Vectorisation, image processing, exports, metadata, downloading and archiving each worked separately. The whole process was a nuisance.

What I built. Browser, API, batch and headless utilities for image and SVG work, tracing, presets, operation history, exports, downloading and archiving. Some make sense alone. Others gradually became one production pipeline.

Use. Used in my own art practice. Small parts while small parts are enough; a system when they need to work together.

SELECTED PROJECTS, NOT A COMPLETE ARCHIVEMORE MACHINES AT CSTM•WRKZ →

03 / HOW I WORK

The real workflow is usually more interesting than its diagram.

See the work.

What people actually do, what they work around, and which inefficiencies have become invisible through repetition.

Shrink the problem.

What must stay, what can disappear, and what does not need software at all.

Build the minimum.

If a script is enough, do not build an app. If an app is enough, do not build a platform.

Test and connect.

Real data, edge cases and the user show what works. More parts only appear when the workflow actually needs them.

AI with boundaries.

I use coding agents and language models heavily, but not as the default answer to everything. Where an exact rule exists, I use the rule. Where privacy matters, I prefer local processing. Outputs that matter must remain checkable.

A model can write a lot of code. It cannot decide for me whether we are building the right thing.

04 / ABOUT

About.

I am a lawyer by education. Today I work somewhere between operations, software, automation and AI-assisted development.

Law taught me to work with rules, evidence, responsibility and edge cases. Administration and business operations showed me how far real work can be from the way it looks on paper.

I build software heavily with current coding agents and language models. My job is mainly to understand the problem, define the boundaries of the solution, connect the right parts and verify that the result works in reality.

Many of my tools start with my own work. If I repeat something often enough, I start asking whether I should be doing it by hand at all.

Technology interests me most when it removes a real problem for someone.

05 / CSTM•WRKZ

CSTM•WRKZ

Most of the smaller machines live next door. CSTM•WRKZ is my technical workshop: small tools, custom systems, pipelines and things that had to exist because general-purpose software was not the right answer. Useful software does not need a large market. Sometimes one person, one team or one unusually specific workflow is enough.

CUSTOM SYSTEMS. TINY MACHINES.

Open CSTM•WRKZ →

06 / CONTACT

“That is just how we do it” is a pretty good place to start.

You do not need to know whether you need AI, an application or automation. You do not need a technical specification either. Describe what happens today, who does it, which part is annoying and what happens when something fails.

Then we can work out whether something should be automated, built, connected or simply done differently.

I am open to focused projects and to a long-term internal role when the problem is more interesting than the job title.

filip@filipbarcik.cz

Understand the work first.Build second.