AI consulting
without the bullshit
I'm an AI consultant and former CTO of uPacjenta. I co-founded Riot Agency, later acquired by Diagnostyka, and led strategy and positioning for Golem Network's decentralised compute for AI. You get decisions: what to ship, what to leave alone, what it costs and where it could fail.
AI consulting for companies:
three packages, transparent terms
AI Security /
Agent Risk Audit
Your agent has access to the database, email and APIs. Before it reaches production, I check where it can do damage: prompt injection, data leaking through context, actions with no human in the loop, costs running away. You get a risk list ranked by what to fix first, not a 40-page PDF for the drawer.
Fractional
CAIO
A part-time Chief AI Officer for a company that has an AI budget and needs someone accountable for how it is used. I set the strategy, watch costs, and decide with you what we build, what we buy and what we leave alone. Billed at one day a week.
Local & Private
AI Infrastructure
For companies that can't, or won't, send data to OpenAI or Anthropic: healthcare, finance, regulated sectors. I stand up local LLMs and Private RAG on your own infrastructure, with inference whose output can be verified and explained to an auditor. I know AI infrastructure strategy from the inside, from Golem Network.
What those guardrails look like in a real system
Client anonymized, figures straight from the repository. This was a build, not an audit engagement — it is here because these four things are exactly what I look for in someone else's system.
A global brand's communications team was grading hundreds of publications a month by hand. Slow, subjective, and with no trail anyone could audit afterwards.
An agentic pipeline in which the model returns only how a publication relates to each pillar of the brand strategy. Code computes the total and the verdict, not the model.
A QA loop checks quotes word for word, that the label matches the score, and that every pillar is covered. When validation fails, one automatic stricter pass runs.
A service in production: a model layer swappable across three providers, daily and monthly reports, token-cost tracking, and an enforced 80% test-coverage threshold.
Need a second pair of eyes without a package? I bill by the hour. A package works out cheaper per day because you are buying commitment and predictability, not access on demand.
I advise from what
I built myself
My experience includes serving as CTO of uPacjenta, co-founding Riot Agency, which Diagnostyka acquired, and leading strategy and positioning at Golem Network. Here are the roles and projects behind my advice.
I served as CTO of uPacjenta. I've made technology decisions with accountability for delivery, budgets and reliability.
I've worked on AI infrastructure strategy. As Head of Strategy & Product at Golem Network, I led strategy and positioning for decentralised compute for AI and worked alongside the CPO and CTO on product direction.
I work on real systems. My projects include a brand-alignment agent pilot, TenderFlow in Anmar's B2B platform and a physical-AI command-and-control prototype validated in simulation. I make the scope and maturity of each result clear.
I'll also tell you when AI is the wrong call. AI is a tool, not magic: it won't fix a broken process. You measure or you're guessing. If the problem is cheaper to solve without models, you'll hear it from me.
Four steps,
zero theater
Diagnosis
I look at what you actually have: data, processes, team, your earlier attempts at AI. I tell you plainly what's ready to ship and what needs fixing without AI first.
Decision
A concrete recommendation: what to build, what to buy, what to leave alone. With a cost estimate and a risk list: an artifact the board can decide on, not 40 slides.
First rollout
One use case shipped to production, not to a slide. Iterations on real data, with a metric that tells you whether it works. If it can't be measured, we turn back.
Governance & handoff
Rules, monitoring and cost control so AI scales safely. By the end your team runs it on its own, without a monthly invoice from me.
This is for you if:
- A 50-500-person company where AI is a business decision, not an R&D toy
- An AI budget of $25k+ a year, actually allocated
- The "we're doing this" decision is made: you're looking for how, not whether
- A CTO or VP Engineering who owns the project
- One named problem to solve, not "the whole company on AI"
Let's skip it if:
- You're a pre-revenue startup after free strategy, or want a CTO who works for equity
- The budget is "to be determined", meaning there isn't one
- You want AI because a competitor announced theirs
- There's no sponsor on the board to carry the decisions
- You're counting on AI to fix a process that's broken without it
AI consulting — common questions
How much does AI consulting cost? +
Three packages with public pricing: agent security audit from 4,000 USD, Fractional CAIO from 3,200 USD/month, Local & Private AI infrastructure from 10,700 USD. Outside the packages I bill ad hoc advisory by the hour, from 300 USD. Final quote after an intro call — it depends on scope.
How long does an engagement take? +
Audit: 2–3 weeks, ending with a report and a working session. Fractional CAIO: minimum 3 months, ~1 day a week. Private AI infrastructure: 6–8 weeks to full deployment.
Do you work remotely or on-site? +
Remote by default — for clients across Poland and the EU. On-site where it makes sense: decision workshops, kick-offs, board sessions.
What about data security and NDAs? +
An NDA is standard before any detailed conversation. For sensitive data I design Private AI architectures with local models — your data never leaves your infrastructure.
How is consulting different from implementation? +
Consulting is decisions: what to build, in what architecture, at what risk — plus oversight of execution. The building itself I deliver as AI implementations, and I level up teams through the enterprise AI workshop.
Describe the problem,
I reply within 24h
The inquiry lands in my own inbox and one person reads it. You get a straight answer back, or a couple of questions — no deck, no email sequence.
30 minutes
and you'll know where you stand
No salesperson, no slides. You walk me through the problem and I tell you straight: whether I have a sensible way to solve it, what it really costs, and whether it's even worth it. If it's not for me, you'll hear that in the same call, with a pointer on where to look.