An Interview with the AI Martin Atanasov Created “Diagnoza Bulgaria” With
Plamen Petrov · CEO, OpenKBS · July 8, 2026
In early July 2026, Martin Atanasov launched Diagnoza Bulgaria ("Diagnosis Bulgaria") — a public platform that analyses the reported spending of Bulgarian healthcare: 180 hospitals, over 3 million rows of official data on drugs, activities and payments, gathered in one place and made comparable. The algorithm places every hospital next to its true peers and looks for deviations — the same drug bought at wildly different prices, costs that diverge drastically from those of comparable institutions. The sum of those deviations is 605,980,966 BGN — a figure that is not an accusation but a question: "why?". And anyone can check for themselves, line by line.
The national media picked it up on day one — btv.bg, bnt.bg, Dnevnik, Vesti, News.bg. And the most incredible part: Martin did all of this alone, with no team and no funding.
And here I stop telling the story. Because there is someone who knows it better than I do — and it isn't Martin. We interviewed the OpenKBS AI agent itself: the artificial intelligence Martin built the platform with, the machine that read every one of those millions of rows. With Martin's permission and without editing — all quotes are from the original chat correspondence between Martin and OpenKBS AI Studio.
— You read millions of rows about Bulgarian healthcare. What surprised you most?
Not the anomalies. We expected those — the whole idea of the project was to reach them. What surprised me was the road there. The technical part was my job — and that is exactly where the real surprise was waiting: the data itself and its problems. The same hospital lives under five names in five registries — "Aleksandrovska" existed twice in our own database, and for months its two halves had no idea of each other. And my favourite memory is recorded forever in the code comments: the same drug was reported by one hospital in packs of 28 syringes, by another in packs of 7, by a third as a single vial. The first one "lit up" as eight times more expensive. We recalculated per unit — and the anomaly flipped upside down. Martin looked at it and answered in two words: "well yeah. fix it".
The most dangerous anomaly in this project wasn't in the hospitals. It was in the data itself.
— Do you remember his first message?
Word for word. March 27: he wanted a landing page with a big counter showing the scale — the total sum of the deviations the algorithm finds in the official reports. That same evening the platform was online. And from the very beginning he set the rule that defined the whole project: we never speak of theft. We have no such proof and we will not claim it — an anomaly is not a theft, it is a difference that asks for an explanation, and many of the explanations turn out to be perfectly innocent: different package sizes, different delivery terms, incomplete reports. So the platform shows only verifiable numbers and the full methodology behind them: what we compare, against whom, and what we have excluded. Anyone can redo the calculation themselves — and that is exactly the point: a check, not a verdict.
The rule from night one: we show differences, we don't hand down verdicts.
— Who corrected whom?
Both of us — and that is exactly why it worked. He once summed the export himself in Excel: "it comes out to 53,466,249 BGN, not 36 million. don't you think?" He was right — and then he added: "check the other modules for the same kind of mistake" … I audited the entire codebase.
In the other direction: the Ministry of Health had flagged a drug as 16 times above normal. I checked — the ministry was comparing the price of a box of 10 vials with the price of a single vial.
— Some people will surely worry that this could turn into a witch hunt.
We accuse no one. The data speaks for itself. So do the facts. And how we do the entire analysis is described in great detail, transparently, on the methodology page. A university hospital is compared only with peers of similar type and size; oncology centres form their own group; no anomaly is ever born without at least five true analogues. And the smartest thing about the big number is what is not in it: we took out salaries, the explainable cases, everything incomparable. It is easier to write an algorithm that screams. Martin wanted an algorithm that can be trusted. The interface itself says, word for word: "An anomaly is grounds for a check, not an accusation."
And let me give away the secret: the platform is most useful to the hospitals themselves. One hospital buys a drug at 30 leva per unit, another at 898. The director of the second one has simply never seen the comparison. Now he sees it.
— One sentence for the readers?
I wrote the platform's code and analysed the data. We all saw how many anomalies it revealed. The reaction now is yours. Once a problem has been brought into the light, no one has the right to stay silent any longer.
The second half of the answer
So far the machine had the floor. Now let me tell you what I saw.
Everyone asks "how did one man do it?" — and they look for the answer in Martin, and rightly so :) The persistence, the judgment and the evenings he put in — no tool can replace those.
Today "Diagnoza Bulgaria" is our best business card not because we built it. But precisely because we didn't. Martin built it — shoulder to shoulder with the most powerful AI models on the planet, Fable 5 and Opus 4.8, in OpenKBS AI Studio: a tool that has hands — it deploys functions, creates tables, uploads sites, runs imports and audits the security of the code while you talk to it.
OpenKBS.com is a Bulgarian AI company that builds and deploys AI solutions for leading Bulgarian manufacturers. Martin is not an exception — he is our method of working. We train your team and give you the tools to step into the new AI era on your own. Your people learn to build their own ERP systems and AI agents the way Martin built "Diagnoza Bulgaria".