Vision systems,
graphics, and the
hard software between.

Real-time systems, machine learning and scientific computing, from prototype to production. Software and engineering consulting for work that matters.

START A PROJECT SEE THE WORK

CAPABILITIES

These days the work sits mostly in these areas:

Graphics & low-level softwareSoftware that works at the level of the hardware itself, and graphics that draw in real time.Read the detail Machine learningTeaching a computer to recognise things: what's in a picture, what someone said, what a document says.Read the detail Web & product developmentThe thing people actually use: the site, the app, the system your staff log into every day.Read the detail Data & analysisAn answer you can act on, out of the data you already have: rankings, forecasts, and patterns nobody has had time to look for.Read the detail

Graphics & low-level software

Software that works at the level of the hardware itself, and graphics that draw in real time.

Most of the work here sits in the gap between what a machine can do and what it is actually doing. In practice that means writing the thing rather than reaching for an engine: choosing data structures the hardware likes, keeping work off the critical path, and being able to account for where every millisecond of a frame went.

Custom renderers in Vulkan and C++20, down to the BVH and the material model.
Real-time GPU pipelines, and the maths that has to be right before anything appears on screen.
Embedded systems down to the circuit board and the electronics around it.
Cases where the ordinary assumptions do not hold, such as light travelling through curved spacetime rather than in a straight line.
Performance measured and profiled rather than guessed at, because the slow part is almost never the part that looks slow.

VULKAN · C++ · EMBEDDED · GPU

Seen in

Relativistic path tracer

Machine learning

Teaching a computer to recognise things: what's in a picture, what someone said, what a document says.

The model is the easy half. The work is the loop around it: what data exists, how it is labelled, what counts as a correct answer, and how you tell whether a change made the thing better or only luckier. Everything is judged on data the model has not seen, because a number arrived at any other way is not a measurement.

Vision models that run on live video without falling behind the broadcast.
OCR and speech to text where the input is messy and the answer still has to be right.
Models fitted on what came before a date and tested on what came after it.
Getting a model out of a notebook and to where the data actually is, which is rarely a laptop.
An honest answer about when machine learning is the wrong tool, because a rule or a query often settles the same question for less.

VISION · LLM · OCR · SPEECH

Seen in

Miekkailusofta

Web & product development

The thing people actually use: the site, the app, the system your staff log into every day.

The interesting part of a product is rarely the interface. It is the model underneath: what a thing is, what can be true about it at once, and what still has to hold a year from now. Get that right and the screens are close to easy. Get it wrong and no amount of front end will rescue it.

Whole products rather than front ends: data model, API, interface and the machine it runs on.
Tools built around the workflow that already exists, not the one that would be easier to code.
Servers, deployment and monitoring after the system is in production.
Accounts, permissions and who can see what, settled before the first user signs in rather than after.
Backups, logs and updates, because a product nobody can maintain is not finished.

WEB APPS · APIS · DATABASES

Seen in

Data & analysis

An answer you can act on, out of the data you already have: rankings, forecasts, and patterns nobody has had time to look for.

Data rarely arrives clean and questions rarely arrive well posed. Half of this work is turning a vague question into one that has an answer at all. The other half is making that answer reproducible: same input, same number, six months later, run by somebody else.

Pipelines that turn source data, however messy it arrives, into something you can query.
Rating and ranking systems: Elo, Glicko and neural models over years of results.
Optimisation and constraint solving, such as a timetable that satisfies the hard rules before the preferences.
Uncertainty presented as carefully as the number itself, since a point estimate with no error bar is an opinion.
Results in a form you can use, a view or an interface, rather than a one-off run on somebody’s laptop.

PIPELINES · MODELLING · RANKING

Seen in

…and circuit boards, electronics, flow simulation and servers when a project needs them.

SELECTED PROJECTS

The live readings for one monitored power supply in UPS Status: 216 V in and out, 1.1 A of current, a 195 W load and 50.3 Hz, each in its own panel.

Grassmark Oy

UPS Status: power outage alerts

A dashboard that watches UPS units over SNMP and raises the alarm the instant one drops to battery, turning a silent power loss into minutes of warning.

MONITORINGSNMPREALTIME

Read the case study

Own product

KurssiSysteemi

A course planner for Finnish lukio students: a conflict-free multi-year timetable built from the real catalogue, around each student's own constraints.

WEBOPTIMIZATION

Read the case study

Every project in full, from a fencing ranking engine to a relativistic path tracer.

All projects

HOW IT WORKS

How a project goes.

01
Talk
Thirty minutes, free. Tell us what you are trying to do and what makes it hard. We will tell you whether we are the right fit for it.
02
Prototype
A small, tightly scoped first step: a first working version you can try yourself, usually within a couple of weeks. Both sides see early what they are getting into.
03
Build
A written progress report every week: what got done and what comes next. You see progress and can steer the direction as it goes.
04
Launch
Code, docs and credentials are yours. No lock-in: continue with us or without us.

ABOUT

From a willow meadow.

Pajuniitty is the farm I grew up on, and where I learned that work is done properly or not at all.

The full story

CONTACT

Start with thirty free minutes.

Tell us what you are trying to do and what makes it hard. You get an honest answer about whether we are the right fit, usually within two days.

START A PROJECT

Vision systems,graphics, and thehard software between.

Capabilities

Selected projects

UPS Status: power outage alerts

KurssiSysteemi

Talk

Prototype

Build

Launch

From a willow meadow.

Start with thirty free minutes.

Vision systems,
graphics, and the
hard software between.