Artificial intelligence is experiencing growing pains—despite record investments, most AI solutions remain isolated in their industry “cages.” Fintech algorithms don’t work in healthcare, medical systems are useless in retail, and transportation technologies don’t transfer to education. This fundamental limitation creates a paradoxical situation: companies are forced to “reinvent the wheel” for each industry, increasing development costs and slowing digital transformation. Meanwhile, the revolutionary “domain-agnostic AI” approach—creating algorithms capable of easily adapting to different sectors, similar to how a Swiss Army knife is effective in various situations—could radically change the rules of the game.
Dmytro Afanasiev was one of the first to transfer AI solutions between industries as different as dentistry and the maritime industry. His recently patented system for optical recognition of seafarers’ documents, which accelerates processing by 70%, initially evolved from algorithms for medical documentation. His company, Seamensway, focused on seafarer training and certification, demonstrated phenomenal growth of 3500% in its first year of operation. Meanwhile, his Tweendeck platform has seen its pilot user base grow by 140% in just the last three months, increasing from 5 to 17 users, and is built on the same principles once used in inventory management for dental clinics. Where do these surprising connections between different industries come from?
How can an algorithm that doesn’t know the difference between dental fillings and ship engines work effectively in both contexts? And why, according to Afanasiev’s methodology, could agriculture become the next beneficiary of this universal approach? In this exclusive interview, the technology entrepreneur reveals his methods for finding basic functional patterns that break through barriers between industries and create truly universal innovations.
Dmytro, in recent years, we’ve observed a trend toward highly specialized AI solutions for specific industries. However, your approach could be called the opposite—you successfully transfer technologies between sectors as different as dentistry and the maritime industry. What is the essence of your methodology for creating “domain-agnostic AI”?
Most developers fixate on industry specifics, while I look for universal patterns. My focus is on the function, not the industry. Document verification is document verification, regardless of whether it’s a maritime certificate or a medical license.
I always start at the ground level—I sit next to a recruiter or clinic administrator and observe their routine. That’s the only way to identify real problems. I remember watching people in a recruitment agency literally drown in paper documents and faxes. The best solutions are born in these pain points.
The methodology itself is simple: break down the problem into elementary functions, find basic algorithmic templates, and then adapt them to the industry’s specifics. It works every time.
In 2016, you implemented a maritime recruitment optimization system at CRS that reduced candidate search and preparation time by almost 50%. Later, you applied similar algorithmic approaches in dentistry. What universal data patterns did you discover between these completely different fields?
You know what’s surprising? In both recruitment and dentistry, everything boils down to a matching problem. A vessel needs a seafarer with specific experience on a certain engine type—for a complex medical procedure, you need a doctor with specific skills. Mathematically, it’s the same problem.
The second pattern is data verification. People embellish their resumes, and patients forget to mention allergies. In the maritime field, we compared a seafarer’s experience with a database of vessels: if they write that they worked on a vessel that didn’t exist during that period, it’s an obvious red flag. We adapted the same algorithms for verifying medical certificates.
I didn’t invent these patterns—I just started seeing that identical mathematical models often hide behind different terms and processes. It’s like in music—different instruments, but the basic harmony is the same.
Your Dentist24.online system, created in 2017, used AI to predict material consumption in dental clinics, reducing inventory by 20-30%. How did you adapt these forecasting algorithms to work with human resources?
We literally replaced consumables with people. In Dentist24.online, the algorithm worked like this: “40 fillings planned → this much composite needs to be ordered.” In maritime recruitment: “15 tankers setting sail → seafarers with such-and-such qualifications are needed.”
Of course, we had to make adjustments. People are unpredictable—they can get sick, change their minds, or find another job. We had to add probability coefficients and extend the planning horizon from three months to a year.
But the algorithm’s core—analyzing historical data, identifying seasonal patterns, and predicting future needs—remained unchanged. Amazingly, the algorithm doesn’t “know” what it’s working with—dental fillings or ship engines—it just looks for patterns in numbers.
Your paper “The Impact of Blockchain Systems on the Transparency of International Logistics Operations” is about to be published in a Scopus-indexed journal“, you explore how blockchain technologies enhance data integrity across global maritime logistics chains. Which methods from this research have you successfully transferred to other industries?
This paper represents my continuing exploration of transferable technologies. The blockchain verification framework we developed for maritime documentation has proven remarkably effective when applied to healthcare record-keeping systems, where data integrity is equally critical.
The consensus mechanisms we designed for validating seafarer certificates across multiple regulatory jurisdictions translated seamlessly to verifying professional credentials in legal and financial services. Both domains require trust in documentation across organizational boundaries.
Perhaps most surprisingly, our method for creating immutable audit trails of maritime cargo transfers has been adapted for tracking patient treatment histories in healthcare settings. The mathematical principles of creating tamper-proof sequential records are identical, regardless of whether we’re tracking containers or medical procedures.
These cross-industry applications further validate my approach to technology development. As I often say, good algorithms don’t care about the domain—they care about the pattern. Blockchain is particularly domain-agnostic because it focuses on the universal problems of trust, verification, and transparent record-keeping that exist in virtually every industry.
Recently, you filed a patent application for a system of optical processing, verification, and unification of seafarers’ documents, which accelerates the process by 70%. How difficult was it to transform experience with medical documentation for application in the maritime field?
I thought it would be easier. I naively assumed: a document is a document, what’s the difference? But the devil is in the details. Medical records are standardized, while maritime certificates are complete chaos—different countries, formats, and languages. Adding to the complexity, the same maritime documents often have different names depending on the issuing authority, and there are various maritime registries where document legitimacy must be verified.
Nevertheless, the basic principles turned out to be identical. Any document consists of semantic blocks. Just as a medical record has a diagnosis and recommendations, a maritime certificate has personal data and qualifications.
The key discovery was contextual interpretation. The term “class” in a medical document means a disease category, while in a maritime one, it means the level of a vessel. Our system learned to distinguish these nuances by context.
The most challenging part was collecting data to train the model. We had to label thousands of documents manually, but the result was worth it. The system now works with 95% accuracy even on previously unseen formats.
You’ve worked with a variety of data in completely different contexts. Which metrics and indicators do you consider truly universal for evaluating the effectiveness of AI solutions regardless of the specific industry?
I stick to three simple metrics: time, accuracy, and money. If AI doesn’t reduce task completion time by at least 30%, doesn’t decrease the number of errors, and doesn’t save or generate additional money—it’s useless, no matter how cool the technology is.
I’m skeptical about vague indicators like “user satisfaction level.” That’s too subjective. In business, numbers matter. Good AI should pay for itself in 12-18 months. Otherwise, it’s just an expensive toy.
Interestingly, the values of these metrics can vary significantly from industry to industry, but the evaluation logic is always the same. It’s like translating between languages—different words, similar grammar.
Your approach to determining an industry’s “information maturity” has helped you identify digital transformation opportunities in medicine and logistics. Which industries, in your opinion, are now on the threshold of an “information breakthrough”?
“Information maturity” is my internal heuristic: how much data is generated, how structured it is, how quickly it updates, and—most importantly—whether it’s used for decision-making.The ideal candidate for a breakthrough is an industry with a large amount of unused structured data. Right now, that’s agriculture. IoT sensors produce terabytes of information about soil, plants, and weather, but 90% of this data sits unused.
Construction is another example. BIM models contain enormous volumes of information, but only a small portion of their potential is used. This data could optimize the entire chain from design to operation.
Education is also approaching. EdTech platforms know more about students than the students themselves: when a person loses concentration, which topics are more difficult, and which material format is more effective. However, this information is hardly used to personalize education.
Your Tweendeck platform, which has been in development since 2023, is currently in its pilot phase with a growing number of test users. You’ve designed this system to account for the entire lifecycle of a maritime specialist from certification to hiring. Which design principles for such end-to-end platforms do you consider universal?
The main mistake most developers make is that they solve point problems, ignoring interconnections. In Tweendeck, we initially considered the entire journey of a seafarer—from training to career growth.
The key principle is that data should be enriched at each stage. Information collected during training affects certification; certification data enriches recruitment. Each module not only uses data but also makes it more valuable for other modules.
The second principle is process adaptability. If analytics shows a growing demand for a certain qualification, the system should automatically restructure training and hiring processes.
The third is independence of modules with tight integration. The client can start with one module and gradually connect others, gaining more advantages with each step. It’s like a well-designed construction set.
You have repeatedly emphasized that “no one needs technology for technology’s sake.” Why do so many AI projects fail when transferred from one industry to another?
The main problem is falling in love with your own technology. Developers create an algorithm and look for applications everywhere, without asking whether it solves a real problem in the new industry. It’s like with a hammer—when you have a hammer in your hand, everything around looks like nails. Many also underestimate context. A solution that works perfectly in e-commerce can fail miserably in medicine, where the cost of an error is incomparably higher.
The third mistake is ignoring integration challenges. The most brilliant AI is useless if it doesn’t fit into existing systems. I’ve seen many technically impeccable projects die because they couldn’t be integrated into work processes.
And most importantly—the human factor. If your solution requires a PhD to work with it, it’s doomed. Success begins not with code, but with empathy for users. This is actually something I’m exploring in my academic research as well. I’ll be defending my PhD dissertation on this subject in fall 2025, focusing on how technological solutions can be made more accessible and practical across different domains without requiring specialized expertise from end users.
Looking at technological trends in 2025, what new opportunities do you see for creating domain-agnostic AI solutions, and which industries can benefit most from this approach?
Federated learning is a game-changer. It allows training models on distributed data without centralizing it, solving confidentiality problems and regulatory barriers. Imagine a medical model learning from the data of multiple clinics without violating patient privacy.
Multimodal models are another breakthrough. The ability to work simultaneously with text, images, and audio is critically important for universal solutions, since information is presented differently in different industries.
Retail, tourism, and construction industries with pronounced seasonality will benefit the most. These industries have similar data patterns and use the same algorithmic cores for optimization. I see incredible potential in the cross-application of technologies between finance and healthcare. For example, risk assessment algorithms used by banks can be adapted to predict disease risks, and vice versa. The future belongs to universal solutions with rapid adaptation to specific tasks. After all, for an algorithm, there’s no difference whether it predicts demand for loans or for medicines—it’s all numbers in different contexts.
