Below is our recent interview with Julia Lebedeva, Head of Healthcare IT & AI Practice at ScienceSoft:
Q: Could you provide our readers with a brief introduction to your company?
A: ScienceSoft is a US AI transformation and software development company specializing in healthcare, insurance, and investment. We have 37 years of experience in IT and AI and a team of more than 750 professionals.
ScienceSoft has worked in healthcare IT for 21 years and completed more than 150 healthcare projects for providers, healthtech companies, and life sciences organizations. In December 2025, ScienceSoft became a Gold Member of the Healthcare Information and Management Systems Society (HIMSS). This membership gives our specialists additional access to industry frameworks, education, and peer knowledge as they continue strengthening their healthcare IT expertise.
The US is one of our core markets. For example, we recently built a HIPAA-compliant telehealth platform for MindCare Solutions, a behavioral health provider delivering telepsychiatry services across all 50 states. We also developed a mobile app for UNM Health Sciences Center supporting the first statewide US implementation of the WHO ICOPE framework for healthy aging.
Q: What makes you the best choice? How are you unique?
A: What differentiates ScienceSoft is the combination of deep domain knowledge and a pragmatic approach to AI.
In healthcare, we bring domain knowledge directly into the project team through ScienceSoft’s medical professionals and regulatory consultants who work alongside our engineers. That means clients do not have to spend the first weeks of a project explaining basic healthcare processes or why a seemingly small workflow change may create problems for clinicians.
As for our approach to AI transformation, our goal is never to just “add AI” or launch as many pilots as possible. We start with the business problem, such as staff shortages, long wait times, patient access issues, slow charting, and so on. Ultimately, those are the problems we need to solve as technology consultants. We ask where AI can bring measurable business improvements: Can it reduce manual work or operating costs? Speed up a process? Help people make better or faster decisions?
For example, if we estimate that an AI agent would speed up referral intake time from 1–3 business days down to 1–4 hours for a particular hospital, we count that as a working AI use case. But sometimes, we discover that the problem a clinic is trying to solve with AI is really rooted in existing processes or problems with legacy systems and interoperability. In that case, the first move is to solve that underlying issue; without it, AI will just be another fancy tool that doesn’t bring any relief to clinicians.
In healthcare, we need to treat AI risks as engineering problems because the cost of an error here is astronomical. First, there are hallucinations. Our team had to develop an entire five-layer output validation architecture to catch common generative AI errors, such as fabrication, where the model can add a medication that was never in the patient record, or critical omission, where it can leave out a medication that was actually there.
Limits on AI authority are another major concern, since we do not want an AI model, even by accident, to give medical advice or change a health record without a clinician’s approval. And then, of course, there is privacy. To comply with HIPAA and similar regulations, you need to limit as much as possible what data AI can access, mask or anonymize it where possible, control where the data is sent, who stores and processes it, and which business associate obligations apply. These are all things we need to work out for every healthcare AI project before we even start building anything.
We take the same value-first, risk-conscious approach to AI internally at ScienceSoft. We use AI extensively to accelerate our software engineering work and deliver faster, but our engineers remain responsible for the quality of the final product, so everything AI generates goes through mandatory human review.
Q: Any highlights about your latest announcement?
A: One of our recent research pieces looked at where telehealth in the United States is heading. Our prediction is that by 2030, around one in ten US medical visits could take place virtually.
We don’t expect another pandemic-style boom. Telehealth has already settled at a relatively stable level — around 6–7% of US medical visits since 2022 — and we expect more gradual growth from here.
There are large differences between specialties. Mental health remains far ahead in telehealth: in March 2026, 28.5% of all mental health encounters were conducted via telemedicine. Endocrinology ranked second, at about 11%, while specialties such as obstetrics and primary care were still at lower levels but showed modest growth in late 2025 and early 2026.
Perhaps the most important conclusion from our research is that technology itself is no longer the main barrier. About 60% of US hospitals already had a telemedicine solution installed as of April 2026. The bigger questions now are regulatory rather than technological: how telehealth should be reimbursed, whether temporary federal flexibilities will become permanent, and how cross-state licensing barriers will be addressed.
Telehealth is clearly not the right fit for every situation. For example, it can increase access to care in rural areas, but it also requires reliable broadband, which is often unavailable there. At the same time, its value is clear in areas such as mental health or routine follow-ups, where an in-person visit may not always be necessary or viable. So, I expect the next stage of telehealth to be less about replacing in-person care and more about finding where virtual and in-person care complement each other best. In many cases, the future is likely to be hybrid.
Q: What can we expect from you in the next 6 months? What are your plans?
A: AI transformation will remain ScienceSoft’s main focus area over the next six months. We expect our clients in healthcare to keep investing in AI initiatives with two main goals: to prove measurable financial and operational value of AI beyond small pilots, and to build enterprise AI governance foundations (controls, integrations, etc.) that would allow them to expand AI to more workflows if and when it proves its value. So, that is what we at ScienceSoft will also be working on: building AI solutions that deliver operational improvements and solidifying our governance practices. At the same time, we will continue deepening our industry expertise in healthcare, insurance, and investment.
In healthcare, we currently see that providers often see the biggest optimization potential in administrative workflows (patient access, documentation, revenue-cycle management, etc.), so that will be our main interest, although we will also continue working on solutions for diagnostics and clinical decision support.
We are also interested in developing software that facilitates the continued evolution of hybrid care, where virtual visits, remote monitoring, and in-person care work as parts of one connected patient journey rather than as separate services.
Internally, we will expand our use of AI-assisted software engineering. The goal is not simply to generate code faster (although many of our clients are interested in that), but to increase productivity while keeping human control over architecture, verification, security, and quality.
We also plan to continue publishing our predictions and trend watches. AI and healthcare technology are evolving very quickly. We want our research to help distinguish mature opportunities that can deliver value now from technologies that still need more experimentation.
Q: What is the best thing about your company that people might not know about
A: One thing many people do not know is that ScienceSoft’s history with AI goes all the way back to 1989. We were founded as an AI product company, long before artificial intelligence became part of everyday business vocabulary. We became the main development partner of Invention Machine. Of course, our early work was very different from today’s machine learning and generative AI: we developed software for computer-aided engineering that was designed to help people solve inventive problems. It went on to be used by major global companies and eventually powered work at 40% of Fortune 500 companies. Since then, we have essentially grown alongside AI itself, through several generations of the technology.
I think that this influences the way we look at today’s AI boom. ScienceSoft has seen technologies and trends change many times, so we try not to become fixated on AI simply because it is everywhere right now.
The question we need to keep asking is: What useful change will this technology actually create?
Sometimes that may mean automating an expensive process and helping people make faster, better decisions. And sometimes, the answer is that AI isn’t the right solution for a particular problem. I think being willing to say that is an important part of being a trusted technology partner.
