Skip to main content
To KTH's start page

From theory to application: KTH students develop AI systems for clinical information challenges

Published Feb 23, 2026

The two KTH students Smyan Sondur and David Sato, have been accepted into the KTH Innovation one year program with their startup idea OptiMed AI, a solution designed to help healthcare professionals navigate and understand large volumes of medical data. When modern language models became available, they saw the potential, especially in healthcare, where information flows are massive and complexity is high. With backgrounds in medical engineering, engineering mechanics, computer science, and economics, the duo combines technical expertise with a strong drive to create societal impact.

The idea is built on an advanced multi‑agent system in which several specialized AI agents collaborate to analyze, structure, and summarize large amounts of medical records. The ambition is to relieve healthcare staff from the time‑consuming and often stressful task of manually going through long and complex patient histories. The system aims to help doctors and nurses quickly find relevant information, obtain clear summaries, and navigate an information flow that can otherwise easily become overwhelming. Their team also includes co‑founders Mina Makar, a specialist in general medicine working at the emergency department at Karolinska Huddinge, and Ismail Ahmed Ali, a specialist in general medicine working at Capio in Västerås.

This need became clear to Sondur and Sato long before OptiMed AI  took shape. Both have personal experiences showing how difficult it can be when important information is not highlighted properly in the care process. For Sondur, it was his father’s treatment that made the problem obvious:

“My father has had psoriasis his whole life. His medication was recently changed, which caused many side effects. But the connection was never really made. When I searched for it in our system, we could quickly see how symptoms and medications were linked.”

Sato’s motivation stems from a similar insight, but from another life situation:

“My grandmother has fairly advanced dementia. You can’t help but wonder,, if healthcare had a tool like this, could patterns have been identified earlier, allowing for more proactive care?”

Their shared experiences made it clear that the problem was not only staff shortages or workload, but also the structure of information itself. Critical details risk being buried deep within growing documentation, even when they may affect diagnoses and treatment decisions. In many patient cases, the documentation is so extensive that it is practically impossible to fully review.

“A chronically ill patient can have 1,00–2,00 A4 pages of medical records. It’s unrealistic for staff to read everything. Large language models (LLMs) are incredibly good at processing vast amounts of data,” Sondur adds.

“OptiMed AI is designed to meet exactly these challenges,” Sato assures.

It was at the intersection of technological possibilities and these personal insights that the duo found their direction. They are also strongly inspired by researchers like Vladimir Vapnik, known for his work in statistical learning theory and its influence on modern AI.

“We wanted to develop something that actually creates societal value, Sato explains. "When we saw the challenges in healthcare, it felt natural to try to solve them with technology we know can make a difference.”

OptiMed AI – a multi‑agent system that makes medical records understandable

By combining modern language‑model technology with agent‑based architecture, the system will be able to filter, analyze, and present relevant information in ways that support clinical decisions without replacing them.

The system will be able to:

  • summarize medical records at different levels of detail
  • enable direct search in patient histories, even in unstructured data
  • adapt to the user’s workflow and specialty
  • integrate into existing medical record systems to minimize extra work
  • use “deep agents,” advanced AI agents with clearly defined expert roles

These agents are what make the platform particularly powerful. Each agent specializes in a medical area, for example cardiology, radiology, or pharmacology, allowing them to interpret specific types of data with high precision. Multiple agents can also collaborate on complex cases.

“This means, for example, that two intelligent agents in different medical fields can ‘communicate’ with each other and jointly analyze a case, just like a multidisciplinary medical team,” Sondur explains.

With this technology, the OptiMed AI team hopes to contribute to more coherent, data‑driven, and navigable healthcare information, where important connections are not lost and staff gain more time to focus on the patient in front of them.

Regulations, trust, and the EU AI Act – a complex landscape

AI in healthcare is a huge opportunity, but also one of the most regulated technological fields. The rules are necessary, but they also slow down innovation.

“We have a love–hate relationship with healthcare regulations, Sato explains. "Mistakes are unacceptable and that’s incredibly important. But it also means the technology evolves more slowly.”

The team actively works to align with the regulatory framework:

  • testing with doctors and partners
  • quality‑assured clinical functions
  • orbit‑layers where humans always review AI output
  • preparations for the EU AI Act, which is still a “grey zone” in some areas

“The biggest barrier isn’t the technology, it’s trust,” Sato says.

EU’s regulations for medical devices (MDR) also pose significant challenges, as the requirements for clinical evidence, documentation, and traceability are considerably higher. This makes both the development process and the approval phase more complex and time‑consuming.

‘It’s a major cost for startups and can be a long process, as the regulation hasn’t quite kept pace with technological development,’ Sondure adds. ‘A year and a half ago, hardly anyone knew what AI agents were. We need to build trust, both in the technology and in understanding who is responsible if something goes wrong.’”

Smyan Sondur and David Sato.
Smyan Sondur and David Sato at KTH Innovation. Photo: Jelina Khoo

Friends since high school – driven by long‑standing curiosity

It has been over ten years since Sondur and Sato first met in high school. They shared a fascination with mathematics, physics, and problem‑solving-subjects that shaped their academic paths and later their entrepreneurial ambitions. Their interest in technology grew early, and for Sato it began as a way to understand how the world works.

“Interest in math and physics has always been there, says Sato. It started with a general curiosity about how things work and why the world looks the way it does. In high school, that interest deepened, and I realized how powerful these tools are for understanding and building things. Then it naturally continued at university.

Sondur describes a similar journey, but with a strong programming interest already in childhood:

“I’ve been coding since I was ten or eleven. I used to build robots, so it’s definitely a core interest of mine. When AI became more accessible, we realized how much potential it had, especially with all the LLMs. That’s when we began thinking about practical ideas and noticed clear gaps in healthcare documentation that we had personally experienced.”

Their paths led them to KTH, Sato currently studies aerospace engineering along with computer science and economics, while Sondur studied medical engineering and completed an exchange term at ETH in Switzerland. Both agree that their time at KTH shaped their direction, in networking, support from faculty, working with large datasets, and the discipline they now carry into their startup.

“I’ve learned a lot about AI from my education,” Sondur adds. “Saying you studied at KTH opens doors. Our network at KTH has even led to some bachelor’s students joining us.”

“Our programs taught us how to approach problems and look for solutions, which we absolutely bring into what we’re doing now,” Sato says.

They are currently putting nearly all their waking hours, plus a bit more into the project.

Accepted into the KTH Innovation one‑year program – aiming for launch

The startup has already received preliminary support from KTH Innovation, but acceptance into the one‑year program gives them new opportunities. The program is divided into three phases:

  1. Sales and customer understanding – pitch, customer contact, offerings
  2. Investment – capital needs, investor dialogues, strategy
  3. Launch – packaging, market, deployment

They are also part of KTH Collide and meet their business coach regularly. Despite being a small team, they have come far.

“We’re a small team, so we basically do everything together,” Sato explains.
Sondur leads programming and technical architecture, while Sato drives much of the business/operational work-customer dialogues, partnerships, strategies, structure.
“But we constantly support each other,” Sondur adds. “If technical input is needed in a customer meeting, he joins. If we need to discuss product or implementation, we do it together. It’s very much teamwork.”

The future: the Nordics first – the world next

The team aims to establish a strong foundation in the Nordics, but long‑term they plan to expand internationally, especially to the US. Over time, they want OptiMed AI to function as:

  • a baseline for clinical intelligence
  • a tool integrated into all medical record systems
  • a platform for research and disease prediction

They envision a future where AI never replaces doctors, but becomes a natural part of every decision.

“We don’t want to replace the doctor. We want to provide a complementary tool that ensures decisions are made based on the right information.”

Text: Jelina Khoo

Glossary (Technical Terms)

  • LLM (Large Language Model)* – Large-scale language models trained on extensive text data to understand and generate human-like text.
  • Agent* – A program component capable of performing step-by-step tasks, using tools, and interacting with an LLM.
  • Multi-agent system* – Multiple digital agents collaborating and functioning as experts in different domains.
  • Deep agents* – More advanced agents with capabilities for collaboration, reasoning, and tool use.
  • Orbit layers* – Layers of human oversight that ensure quality and accountability in AI‑driven processes.
  • RAG (Retrieval-Augmented Generation)* – A technique where AI retrieves facts from external sources before generating its response.
  • EU AI Act* – The EU’s upcoming regulatory framework governing how AI may be developed and used across different risk levels.
  • Biological medicines* – Medicines derived from living cells or organisms, often highly specialized and effective but requiring specific follow‑up and monitoring.
  • MDR (Medical Device Regulation)* – is the EU’s regulatory framework that ensures medical devices are safe, reliable, and function as intended before they are used in healthcare. The regulation sets requirements for how products are developed, tested, documented, and monitored — both before and after they reach the market. It applies to everything from implants to advanced software and AI‑based solutions, and introduces stricter demands on clinical evidence and clear traceability.