Problem → solution

Solve a hard technical problem

A hard technical problem cannot be solved by standard design: there is no ready algorithm, the data behaves unpredictably, and the outcome depends on accuracy. Here the method itself becomes the subject of the work. We frame a testable hypothesis, research it and take the solution to a working prototype.

For companies and teams that need a result the market does not offer: their own algorithm, a model built for their task, non-trivial data processing, or a test of a technical hypothesis before investing in a product.

What people come with

No ready solution exists

The problem cannot be closed by buying a service or by standard development: it needs an algorithm or a model the market does not have, or one whose accuracy and licence do not suit you.

The data behaves unpredictably

Labelling is inconsistent, formats diverge, the signal drowns in noise. A simple parser or an off-the-shelf model breaks on real data, even though everything worked on the demo examples.

Unclear whether it can be solved at all

There is a technical hypothesis, but it is unclear whether it will hold and what an attempt will cost. Investing in full-scale development blind is risky.

How we solve it

  1. Frame a testable hypothesis

    We turn a vague goal into a hypothesis with a success criterion: what counts as a result and how we will measure it. Without that, research can go on forever.

  2. Research and test early

    A review of the field, experiments, early tests on your data. We take on the research risk and report a negative result at once, before delivery.

  3. Take it to a prototype

    From a confirmed hypothesis to a working solution and tests to an agreed methodology. If needed, we take the prototype to a service in operation.

  4. Transfer the technology

    Source code, documentation, stage reports and team training. The result stays with you; we do not build in a dependence on us.

What it is built on

Algorithms and models for the task

We develop solutions for the task when ready-made ones do not fit on accuracy, licence or deployment requirements. Part of our research work is open: you can download the code and check it.

Formal proofs

Where an error is unacceptable, the properties of a method can be proved formally. For the Balansis model, 153 theorems are proved in Lean 4 with no gaps (as of 05.10.2026). They prove the algebra of the model over the real numbers, not the behaviour of float64.

An honest metric

We compare every result with a simple baseline and state its limits. If the result is negative, you get it early and cheaply, with the reasoning.

What we do for it

What it looked like in practice

Manufacturing · Building materials

MVP in operation

Panel specs from CAD drawings in seconds, not hours

Reads DWG/DXF layouts and builds the Excel spec: 10–⁠30 seconds instead of 2–⁠15 hours by hand. All 784 panels of one real project matched the manual spec line by line.

784/784 and 189/189
panels on one project and wall panels on another matched the manual specs line by line, size groups included; area within 0.001 m²
2–15 h → 10–30 s
to produce a project spec: manual work vs app processing, measured on real projects

Marketplace e-commerce · Sales analytics

In production

Profit per SKU across three marketplaces, checked against manual figures

Five seller accounts on WB, Ozon and Yandex Market merged into one SKU-level P&L. Variance from the client’s manual figures on a control SKU for one month: 0.00%.

0.00%
variance vs the client’s manual benchmark across units, revenue, COGS, profit and margin for a control SKU over a month; the threshold was 1%
1.18M rows
of sales across 5 accounts and 12 months processed; 10.7% were ordered and settled in different months

Scientific computing / R&D

Research project

From mathematical theory to a verified numerical library in 9 months

Compensated-computation library: precision loss is visible and checkable, the model’s algebra is proved in Lean 4, each numeric claim has a reproducible benchmark.

1.0 instead of 0.0
residual preserved on the [1e16, 1, −1e16] sum that float silently drops; reproducible benchmark generated by a script in the repository
6 of 6
scenarios of the published benchmark in which the exact dot-product kernel matched the exact reference; numpy.dot matched in 0 of 6. The C kernel is not yet released on PyPI, and the cancellation scenario was built on purpose

Work it out on your own figures

The AVA calculator works out the economics of your process from your own figures, with no form and no price from us. If automation does not pay off, it says so.

Estimate the payback

Frequently asked questions

Is there such a thing as R&D in software?

Yes. R&D is defined by the nature of the work, and the industry plays no part: there has to be scientific or technical uncertainty that standard design cannot remove. Developing a new algorithm or model for a task with no ready solution meets this criterion.

What if the result turns out negative?

That is a regular outcome of research, and it has value too: you get a reasoned answer that the path does not lead to the goal before major investment. We plan the stages so that such an answer comes as early and as cheaply as possible.

Do you take research all the way to a working product?

Yes, if you need it. One of our case studies is an example: an application that builds sandwich-panel specifications from CAD drawings is in operation as an MVP. On one real project, 784 of 784 panels matched the manual specification line by line.

Tell us what you are dealing with

Describe it in your own words. If it can’t be solved or won’t pay off, we’ll say so before any work starts.

In the reply we will say what we understood, what we need to clarify and whether it is worth doing.

or email us directly: hello@xteam.pro

See case studiesNew write-ups on Telegram