Custom software · AI systems

Custom AI systems, built for small and mid-sized companies.

Large enterprises buy platforms and hire integrators to bend them into shape. A company of twenty or two hundred people needs something else: software that fits the way it already works, built by a team small enough to talk to directly. That is what we build - with AI where it pays off, and plain, solid engineering everywhere else.

Start a conversation See what we built

AI on your dataInternal systemsLogistics

● Incoming message → record

Hello, I need a quote for 2 pallets, ADR class 8, from Gdansk to Rotterdam. Cargo ready next week. Please send an offer.
Regards, Marek Kowalczyk

RouteGdansk → Rotterdam
Cargo2 pallets
ADRclass 8
CustomerMarek Kowalczyk
Pickup datenot in the message

Four fields from one paragraph. The fifth is not there - and the system says so, instead of guessing.

What we do

Three things we do well.

We are not a general-purpose agency. These are the areas where we have working code and can show how something was built.

Scope 01

AI on your data

Systems that answer only from your documents, cite the source of every answer, and say so when there is no source. Structured data extracted from email, PDFs and scans.

How we build it → Scope 02

Internal systems

Applications that replace the spreadsheet and the shared inbox: workflow, operational dashboards, integrations with what you already run. Rarely glamorous, and often where most of the hours are won back.

How we build it → Scope 03

Transport and logistics

The industry our own product was built for. We know what a quote request looks like, what an ADR class changes, and that a forwarder gets forty emails a day with no time to retype them.

How we build it →

In practice

What this looks like in a company like yours.

Most of our work starts with one of these. If yours is on the list, the first call will be short.

The shared inbox

Three people read every email to find the ones that matter.

Messages arrive sorted, with the key details pulled out and a reply drafted for approval.

Supplier invoices

Someone retypes each PDF into the accounting system.

Fields are read from the document; only the ones that do not add up reach a person.

"Ask Anna, she knows"

Procedures live in one long-serving employee's head.

A knowledge assistant answers from your documents and links to the page it used.

The quoting spreadsheet

Prices are built in a file only one person dares to edit.

A quoting tool with your rules, permissions and a history of every change.

"Where is my order?"

Customers phone for status updates your team has to look up.

A simple customer portal showing what your system already knows.

The monthly report

Two days a month copying numbers between files.

A dashboard that is current every morning, with the numbers that need a decision.

These are typical starting points, not a menu. The right first project is whichever one costs you the most hours.

Proof

Enterprise Brain - our own product, up and running.

The most honest proof a software house can offer is a system it built and still maintains. Enterprise Brain reads a freight forwarder's inbox, extracts data from quote requests, and answers questions about company procedures - always citing the document and page.

Principle: AI proposes,
a person decidesBy default nothing reaches a customer without approval. Automatic sending is a deliberate, per-customer choice.
Source: An answer
or "I don't know"With nothing in your documents to back it, the system declines. An invented answer is worse than none.
Trail: Every decision
recordedWho, when, on what basis, using which model. The running code version is visible from inside the application.

What Enterprise Brain does

Your first project

Small enough to start this month.

A first project is sized to prove the idea on your own data, not to transform the company.

Week 0 · 30-45 minutes
A call
You describe the process. We ask where the data comes from and who decides. If software is the wrong answer, you hear it here, not in a proposal.
Typically 2-6 weeks
A pilot on one process
Fixed scope, a measurable outcome and a date. You see it working on your real documents as it is built, not on slides at the end.
Then
Handover, or the next process
You get the repository, tests and run instructions. Maintenance is agreed before we start. Whether we build the next piece is your call.

How we work

One process first, not a transformation.

The biggest risk in an AI project is not the model. It is building something nobody uses because it changes how twenty people work all at once. So we start with one narrow process that can be measured - and with the question of how we will both know it worked.

You talk to the people who build it

No account managers, no hand-offs between sales and delivery. The engineers on the first call are the engineers writing the code, which is most of why small projects with us stay small.

A person stays in the loop

We build systems where people approve rather than clean up after automation. It is also the only arrangement in which you can later show why something was sent to a customer.

You own what we leave behind

The repository, tests, run instructions, and an honest list of what breaks if something is forgotten. A project ends with a working system handed over, not a presentation.

Questions

Before you ask.

Are we too small for custom software?

Probably not. We scope to one process rather than the whole company, so the first project is sized for a business of tens of people, not thousands. And if an off-the-shelf product already does the job, we will tell you so and stop there.

Do you work with companies outside Poland?

Yes. We work in English, remotely, on Central European Time. Data is processed on EU endpoints by default, which tends to simplify the conversation with your own customers.

Is our data used to train models?

No. We design systems so that customer content is not used for training and is processed on endpoints in the European Union. The exact terms depend on the model provider chosen, and we put them in the contract rather than promising them in general.

What if the model gets it wrong?

We assume it will. That is why the systems we build send nothing on their own by default, show the source of every value, and decline to answer without support in your documents. A mistake should be visible before anything goes out, not after a customer complains.

How soon is the first working result?

It depends on the process, but we aim for weeks, not quarters - because the scope is one process, not the whole company. We set the date after the first call, once we know where the data comes from and who makes the decision.

Tell us which process eats your day.

We answer within one working day - including when the answer is that software is the wrong tool for it.

Get in touch