AI consulting · AI workshops
We help startups, and single teams inside larger companies, work out what is genuinely worth building with AI — then we help build it, and we teach your people to keep it running without us.
First call is 30 minutes, no charge. We will tell you plainly if you do not need us.
01 — The actual problem
Models are commodity now. What is not commodity is the discipline around them: who reviews generated code, how context reaches the agent, what happens when a run costs four hundred euros, and whether anyone can explain last Tuesday's incident.
Symptom 01
Team ship code. Nobody can say why there is a new column in the database.
Symptom 02
Throughput is genuinely better, but review load moved onto two senior engineers who are now the bottleneck and the safety net at once.
Symptom 03
Inference cost is an unowned line item. No budgets, no personal attribution, no idea whether the spend buys anything measurable.
Symptom 04
Six impressive demos, zero of them inside a workflow that a customer pays for. The gap is integration and ownership, not model quality.
02 — Services
Consulting or workshop, quoted up front and scoped in writing. No hourly billing, no open-ended retainer you are afraid to cancel.
Fixed scope
Two weeks, one team, one honest document. Where AI actually pays off in your product, what it will cost to run, and what to fix before you build anything.
From €7,500 · 2 weeks
Request the auditOngoing
Most commonTwo to eight days a month alongside your team. We sequence the use cases, design the workflow and architecture, setup agents where it counts, and leave behind standards your team keeps.
From €5,900 / month · pause or cancel anytime
Check availabilityPer session
Ninety minutes to get a whole room onto the same page, or a hands-on day where your team builds something real. On site or remote, in English or Czech.
From €1,800 · per session
Book a lectureIndicative pricing for a Czech or EU-based team of five. We adjust scope, not the hourly rate — there isn't one.
03 — Lectures & workshops
The fastest way to find out whether we are useful is to put us in front of your team for ninety minutes. Not a vendor pitch: a working lecture on what these systems actually do, where they fail, and what it takes to run one you can trust.
Talks we give
What genuinely speeds up, what quietly regresses, and how review load moves. Instruction files under version control, permissions, sandboxing.
Building the small regression set your engineers maintain themselves — and gating a deploy on a exact check instead of a feeling.
Per-team attribution, hard caps, cheap-model routing, caching — and the point at which self-hosting open-weight models actually pays.
Any of these can be cut to a 45-minute version, or combined into a half-day workshop with your own codebase on the screen.
04 — What we deliver
Every consulting engagement leaves files in your repository: prompts under version control, evaluation suites, review checklists. Things your team can argue with and improve long after the invoice is paid.
Repository instruction files, agent permission boundaries, and a shared config your whole team inherits.
What an agent may merge unattended, what always needs a human, and how generated code is treated differently in CI. Written down, enforced automatically.
A regression suite for code changes, so "it feels better" becomes a check you can gate a deploy on.
Per-team attribution, hard caps, model usage, caching and memory.
What the agent should read, what it should, and what it should never see. Using memory for continuous improvement.
A short monthly note on what actually changed in agent tooling, run against your codebase — so you adopt deliberately instead of by conference slide.
05 — Fit
Scope discipline is the whole product. If your problem needs forty people and a programme office, we are the wrong vendor and we will say so on the first call.
four to thirty people, one or two engineers, a product that real customers use, and a founder making architecture decisions at midnight. You need judgement now and someone to teach the team as it grows.
A platform, data or product team of three to twelve with a real budget and a mandate to move faster. We work inside that team's boundaries and leave your enterprise AI programme alone.
Company-wide AI rollouts, model training or fine-tuning at scale, procurement and compliance programmes. Those are legitimate problems — they are just not ours, and pretending otherwise wastes your budget.
06 — How it runs
Thirty minutes. You describe the team and the blocker; we say whether we are useful. Often it ends in a referral — or in a lecture booking instead of a project.
Week 0
We read the code, sit in your standups, and interview three or four engineers. Output is a one-page problem statement you can argue with.
Weeks 1–2
Toolchain, guardrails, evaluation harness, cost visibility. Unglamorous, foundational, and the reason later wins hold instead of evaporating.
Weeks 3–6
One workflow reaches production with tests and runbooks. We reduce days per month as your team takes full ownership. We stay in touch if you need help in the future.
Quarter 2 onward
07 — Questions
Still unsure? Ask directly — the first call costs nothing and we answer awkward questions properly.
Ask a questionA mix of use-case triage, architecture review, hands-on pairing, and writing down the decisions your team keeps re-litigating. On a two-day month that is one deep working session and one review. On eight days it looks like a part-time senior engineer who also owns the roadmap conversation.
Big firms sell a programme and a platform; we sell a short engagement and then leave. We resell nothing, take no vendor commissions, and measure success by how quickly you stop needing us. You keep the artefacts either way.
Yes — a lecture or workshop is a standalone booking, and it is how a good number of longer engagements start. No prerequisite and no follow-up sales sequence. If the ninety minutes were useful, we will talk about the next step; if not, you have still had a useful afternoon.
Whatever fits and is the current best, including self-hosted open-weight models where data residency or cost demands it. You should never be locked in by our recommendation.
Remote by default, with one on-site day per month in Czechia where it helps — whiteboards still beat video for architecture. Lectures and workshops we deliver remotely, on site anywhere in Czechia when that suits the team better. Custom travel agreement is possible.
That is the cheapest moment to have the conversation. The first win is usually deciding what not to build, and a two-week audit is designed for exactly that stage. An introductory lecture is a low-commitment way to get a whole team asking better questions.
09 — Contact
One message, a few lines, no form ceremony. We reply within one business day — usually with either a call slot or an honest "not a fit".