AI-first software company, EU
Opens Calendly in a new tab
We install AI-driven delivery pipelines in your team and run one ourselves: coding agents draft, engineers review, a named person signs every release. We come from financial services and medical-device software. We ship agents, LLM features, adapted models, and on-device inference. We tell you, in writing, what AI should not do in your case.
01Services
When you need a decision about AI, architecture, or both.
Architecture and AI Readiness Assessment, two weeks, fixed scope.
When you need it built, with AI where it earns its place, and kept running.
Discovery Sprint, three weeks, fixed scope.
02What we build
Coding agents, review gates, evaluation suites, and release automation in your repositories and CI, in your tenancy. Every change is traced, from the prompt to the approver.
Task agents with tools, limits, logs, and an owner. Retrieval, extraction, and drafting inside the products you already run, with an evaluation suite that blocks a release when quality drops.
Retrieval design, fine-tuning, distillation, and quantization for your domain, with the evaluation set built first. An adapted model replaces the base only when the numbers say so.
Models that run on the device: quantized, measured for latency, memory, and energy on the hardware they ship on, validated as a component with a signed update path.
Custom platforms and back-office systems. Monoliths split at the seams that matter. Data moved without losing the audit trail.
Our own startups, built AI-first and operated on our account. A practice runs there before we sell it.
03How we build
Our engineers work with coding agents in a loop: research, plan, implement, review. Nothing merges or deploys before a named engineer reviews the diff and the test run. This is the pipeline we install for clients, and it runs on our own products first. We learned the discipline on medical-device and banking projects, where every change must be explained later.
04Where AI stops
Every AI claim on this site is only as credible as the limits stated next to it. These are ours.
Reads a whole codebase and drafts a map a person checks. Drafts code, tests, and documents. Classifies and extracts with a measured error rate. Makes the same bounded change across many files.
Know when it is wrong. Hold a requirement steady across weeks. Give the same answer twice. Tell you that your problem does not need a model at all.
Merge to the main branch. Touch production data. Approve validation evidence. Make a decision a patient, a customer, or a regulator will feel, without a named person signing it.
Every assessment ends with a written verdict: where AI belongs in your case, where it does not, and what stays with people. If the answer is "no model", you read it from us first.
We work with AI every day and keep score. Reviewer acceptance per agent; an agent whose acceptance drops is pulled. An evaluation set on every feature. Your own backlog in every assessment. That is how we know where it stops.
05Method
Solid Ground is our name for the way we run every engagement: four stages, each closed by a gate where you decide whether to continue, change direction, or stop. AI moves work inside each stage; it removes no gate. You never pay for a stage you have not agreed to.
06Our products
Running products is the honest test of a lifecycle story, and of an AI story. These are some of ours, run from the first diagram to the monthly cloud and model bills; the rest are on the Startups page. The stage on each card is the stage today, not the ambition.
A person still decides in every one of them.
StageLive
An AI accountant for small businesses in six EU countries. Invoices and bank statements go in; VAT, income tax, and payroll filings come out as filing-ready XML, for EUR 10 per month.
StageMVP
IT support by subscription for business software that outlived its team. The AI holds the project knowledge and drafts each fix inside a seven-stage maintenance pipeline.
StageLive
Short interactive games paired with essays on the technology behind them: seeded randomness, provably fair validation, and why the house wins.
IT will rock for:
07Insights
Articles on AI principles and delivery. Short, dated, from real work.
Article12 min read
By Artem Panfilov
SOCIA taken to automation: agents hold the gates between phases, read only the contract and the artifact, and send work back until it passes, while a named person signs the gates with consequences. Why it works starts with a human blind spot: we cannot see our own mistakes, and neither can a model reviewing its own window.
Article12 min read
By Artem Panfilov
Separation of concerns, applied to working with AI: one session per concern, a one-page file as the only interface, and a gate a person holds at every seam. Why that structure costs fewer tokens, is easier to govern, and keeps you free to change vendors.
08Facts
Bring one problem to a 30-minute call. We will put a senior architect on it and tell you, in plain English, what we would do, where AI belongs in it, where it does not, and what it would cost. If we are not the right team, we will say so and tell you who is.
Opens Calendly in a new tab