Most companies have already bought AI tools. Far fewer have changed how the work runs around them, and that is where the gains are. I'm Vlad Tudor, an AI engineer in Bucharest and the founder of Sapio AI. I help companies work out where AI genuinely fits, what to build, and what to leave alone.
Reply within 24 hours · no sales pitch
The licences are paid, a few people use them well, and the rest of the process looks the way it did before. Usually the tool isn't the problem. The work around it was never redesigned, and the data it needs is scattered across spreadsheets, inboxes and people's heads.
An idea with AI somewhere in the middle, and a decision to make before spending money on it: what the technology can carry today, what it can't, what to build first and what to buy instead. A second opinion from someone who builds these systems for a living.
Confidentiality and cited sources, from the creator of ai-aflat.ro.
Assistants that answer from official sources, and help with the tender before you buy.
What AI should answer, what stays with a person, and when a ready-made widget is enough.
What to build, what to buy and what to leave out, before money is spent.
Manufacturing, energy, distribution and back office: fix the process, then automate it.
Before electric motors, a factory ran on one steam engine turning a central shaft, and every machine connected to it. When electric motors arrived, most factories simply swapped the engine on the shaft. For decades, the gains were small. They came when factories were rebuilt around the task instead of around the power source.
AI is at the same point in most companies. Adding a model to an unchanged process gives you a slightly faster version of the old process. The work that pays is rewiring the process itself: which steps disappear, which get checked by a person, and what data has to exist before any of it works.
The full argument, from the REWIRED podcast.
About the problem, before any technology. Free.
How the work runs today, what data exists, which tools are already paid for.
What to build, what to buy and what to skip, in order, with the reasoning written down.
Hands-on build through Sapio AI, or a roadmap your own team can take from there.
A workshop for the people who will use the result, so it gets used.
A check on what changed once the work is in place, and what comes next.
One call about the problem, not the technology. What you're trying to change, what has been tried, what a good outcome would look like. Free, and it's often enough to tell whether AI belongs in the answer at all.
A look at how the work actually runs today and what data exists to support it. This is where most AI projects are won or lost: messy data and unwritten rules sink more projects than model choice ever does.
What to build, in what order, what to buy off the shelf, and what to leave alone. Each item with the reasoning behind it, so the decision stays yours and can be defended internally.
If the roadmap says build, Sapio AI, the studio I founded in 2021, can build it to production. If your own team is better placed, the roadmap is written so they can take it from there.
Building AI systems since before ChatGPT made it mainstream. That makes it easier to say honestly what AI can't do.
The person who scopes the problem understands the build. No hand-offs between consultants and developers.
When the answer is to build, Sapio AI builds it, so the plan doesn't stop at a slide.
Conversations, documents and workshops in either language, for teams in Romania and abroad.
A demo that impresses in a meeting and falls over on real data helps nobody. The work is judged by what runs in production.
Sometimes the answer is a better form, a cleaner spreadsheet or a rule written down. You'll hear that, even when it means there's nothing to build.
AI amplifies whatever it's given. If the process is unclear or the data is a mess, that gets said before anyone spends money on a model.
A free assistant that searches the whole of Romanian legislation, picks the laws relevant to a question and answers with the law cited and linked to the official source. Started as a side project in summer 2024; grown with no marketing budget.
vladtudor.com/ai-aflatOne of the first ChatGPT-based customer-support agents in production, built for a retail startup at the start of the LLM era. The company went on to raise a $4.7M seed round.
As an ML engineer, trained what was then the best-performing Romanian speech-to-text model for a client.
Describe the problem in a few short questions. The first conversation is free.
Reply within 24 hours · no sales pitch
Vlad Tudor: AI engineer and founder of Sapio AI (2021), creator of ai-aflat.ro. Forbes Romania 30 Under 30 (2025), TEDx speaker, and four years teaching at Politehnica University of Bucharest. Every engagement is led personally.
Works out where AI can change how your company works, and where it can't. In practice: a conversation about the problem, a short audit of your processes and data, and a roadmap that says what to build, what to buy and what to leave alone. The value is in the decisions, not the slides.
It depends on scope. The first conversation is free, and you get a written proposal before anything is agreed.
It depends on scope too. The conversation is a single call; the audit and roadmap are kept short on purpose, and a build runs as long as the roadmap says it should.
Yes, through Sapio AI, the studio I founded in 2021. It builds production AI systems: assistants and agents, retrieval over documents (RAG), computer vision and speech-to-text. If your own team would rather build, the roadmap is written so they can.
Both. Conversations, workshops and written material in either language.
Either. I'm based in Bucharest, so meeting on-site there is easy; elsewhere the work runs remotely, with on-site days when they help.
A few short questions, two or three minutes. The reply comes from me, within 24 hours on working days.
Reply within 24 hours · no sales pitch
Help with
Pick all that apply.
