Techie.Lk recently had a conversation about Artificial Intelligence (AI), multi sector technology and the future of work with MEDCUBE USA General Manager, Kennedy J. Michael, who has spent many years of his career working at the intersection of business, people and technology, across a number of sectors.
The interview mainly spotlighted the recent collaboration between healthtech brand MEDCUBE USA, XPAND, and the Academy of Artificial Intelligence and Information Technology (AAIIT).
MEDCUBE USA is a healthcare software and technology company that operates between North America and Sri Lanka, with XPAND as its offshore talent sourcing arm, and AAIIT as its tech education provider.
What is the strategic objective driving the collaboration between AAIIT’s talent development pipeline and MEDCUBE USA’s technology ecosystem?
MEDCUBE USA is involved in healthcare, legal case management, document summarisation, document management, billing and other information intensive processes.
These may look like different areas, but they all have one thing in common: they involve people dealing with large amounts of information. This is where I see a tremendous opportunity for AI.
For me, AI is not about chasing behind the latest technology. It is about understanding a business problem and identifying whether technology can solve it better, faster or more intelligently.
Having worked across different sectors, I have seen the same problem repeatedly: what people learn and what businesses actually need are not always the same.
This is one reason why we created AAIIT. I don’t want it to be an institute where people simply collect certificates. I want people to leave understanding how technology is actually used in businesses.
At AAIIT our objective is to help people understand how technology fits into real businesses, not to simply teach them how to use AI tools.
We want to build professionals who can walk into an organisation, understand a real problem, use technology intelligently and create real value.
So for me the connection is quite simple. AAIIT develops people with practical skills, MEDCUBE USA gives them exposure to real problems, and XPAND can help connect that capability with international opportunities.
I have always believed that education should lead to capability, and capability should lead to opportunity.
How does integrating AAIIT’s applied AI work into MEDCUBE USA’s operations enhance product capabilities or operational workflows?
I have never believed in using technology simply because it is available. The starting point should always be the problem.
If we look at our own businesses, we deal with large amounts of information every day. In healthcare it may be medical records, referrals or billing information. In legal technology it may be case files and correspondence. In document management it may be hundreds of documents that someone has to read, classify or retrieve.
AI can help with some of that work. It can extract information, identify missing details, summarise documents and help people organise their work.
But I would always measure the result. Did we save time? Did accuracy improve? Did we reduce rework? Did the staff actually find it useful?
This is how I look at technology. It has to improve the business, not just make a good demonstration.
Where do you see the biggest potential for AI adoption in healthcare and other information intensive operations?
I think the biggest opportunity in the short term is in the work done around the professional, rather than trying to replace the professional.
Healthcare is a good example. There is a tremendous amount of administrative work around patient care. There is also a similar pattern in legal services, where professionals spend a great deal of time going through documents and extracting information.
AI can help with document classification, summarisation, transcription, information extraction, referral checking and workflow management. These are areas where you can make people considerably more productive.
My view is that AI should take away some of the repetitive work so people can spend more time on judgement, communication and decision making.
How do you measure the long term Return on Investment (ROI) and impact of the AAIIT MEDCUBE USA relationship?
I am a great believer in measuring outcomes rather than activity.
It is easy to say that we have introduced AI, trained people or developed a new system. The real question is whether any of it made a difference.
For the business, I would look at turnaround time, accuracy, productivity, rework, customer satisfaction and cost.
For AAIIT, I would look at whether our students are genuinely work ready, whether they can solve practical problems and whether they are getting meaningful opportunities.
Commercially, I would look at whether we are building sustainable products, clients and international opportunities.
Over three to five years, I would want to see evidence that technology is improving the business while people are becoming more capable. To me this is real ROI.
What specific AI or Natural Language Processing (NLP) technologies are being used to process unstructured healthcare, legal and business documents?
I am careful about answering this question simply by listing model names. Technology changes very quickly, and naming a particular model does not prove that you have solved the business problem.
Depending on the use case, we can use Optical Character Recognition (OCR), document extraction, language models, classification, summarisation and rules based validation.
But I always start with the use case.
What are we trying to achieve? How accurate does it have to be? What happens if it is wrong? How do we test it? Who checks the result?
We are working in areas such as healthcare and legal case management where an attractive looking answer is not enough. The output has to be useful and verifiable.
This is why I am more interested in building good AI enabled processes, rather than simply saying that we are using the latest model.
How are you addressing AI hallucinations and errors?
This is one area where I am quite cautious.
If an AI system produces a wrong answer in a general conversation, it may be inconvenient. If it produces the wrong information in a medical record, a legal case or a financial document, the consequences can be much more serious.
So we have to design the process around that reality. The system should work from authorised information, the output should be capable of being checked, and there should be clear limits around what the AI is allowed to do.
I don’t believe that a confident AI answer should automatically become a fact.
For higher risk work, there must be human review and clear accountability. This is also why I keep coming back to the idea of Compliant AI. AI not only has to be intelligent, but controlled, traceable and appropriate for the environment in which it is being used.
What approach should be taken to AI lifecycle management as healthcare, legal and business requirements change?
One lesson I have learned from software and business operations is that change is constant. Regulations change, customer requirements change, processes change and software changes.
So AI cannot simply be developed once and forgotten.
We need controlled versions of the models, prompts, business rules and reference information. We need proper testing when something changes, and we need to monitor performance over time.
Importantly, not every change requires retraining an AI model. Sometimes the right answer is to update a rule, a reference source or a workflow.
The important thing is discipline. We need to know what changed, why it changed, what was tested and who approved it.
Do you see value in proprietary domain specific AI models for Sri Lanka and the region?
I would not build a proprietary model simply so that we can say we have one.
There are capable foundation models available. I would rather take those technologies and apply them intelligently to problems we understand well.
Our opportunity may be in the surrounding intellectual property, the workflows, document structures, domain rules, validation methods and integrations that make AI useful in a particular industry.
Healthcare, legal services and document management all have their own requirements. There may also be opportunities for regional languages and local workflows.
If there is a genuine commercial or technical reason to develop something proprietary, then we should do it. But the starting point should always be a real problem and a real market need.
How do you approach privacy, security and compliance when AI is used with sensitive information?
For me, privacy is not an IT checklist. It is a business responsibility.
When you work with healthcare information or legal case information, people are trusting you with something important. That trust can take years to build and very little time to lose.
We therefore need to understand what information we are handling, who should have access to it, where it is stored, how it is protected, how it is shared and how long it should be retained.
Where applicable, we have to work within the relevant legal and contractual requirements, including healthcare privacy requirements such as the Health Insurance Portability and Accountability Act (HIPAA).
And we have to be sensible about AI tools. You cannot simply take sensitive information and put it into a public AI service without understanding what happens to that information.
That is the foundation of compliant AI: technology that operates within the rules, controls and responsibilities of the business.
How do you approach synthetic data and anonymisation for AI development and training?
I think this is an area with enormous potential for both education and product development.
We need realistic problems for people to work on, but that does not mean giving students or developers access to real patient or confidential legal information.
We can create realistic synthetic cases, referral forms, documents, transcripts and billing scenarios, including the messy situations that occur in real life such as missing information, inconsistencies and exceptions.
That allows us to test systems and train people without unnecessarily exposing sensitive information.
But synthetic data still needs to be handled responsibly. We should not assume that anything labelled synthetic is automatically safe.
Again, it comes down to good governance.
What governance should be in place before AI is deployed into business critical workflows?
I don’t think every AI application needs the same level of governance.
If an AI system is helping someone organise appointments, the risk is different from an AI system that is being used around medical information, legal case decisions or billing.
So I would start by identifying the risk of the particular use case.
There should be someone responsible for the system, a clearly defined purpose, acceptance criteria, security and privacy checks, proper testing, an escalation process.
We also need to know when the human being has to take over.
I would rather take a little longer to put the right controls around a system than move quickly and discover later that nobody knows who was responsible when something went wrong.
How do you see AAIIT students and researchers working with real industry challenges?
This is one of the areas I am particularly interested in.
People learn much faster when they are working on something that resembles a real business problem.
We can give students practical challenges based on healthcare, legal technology and document management. For example, building a document test suite, evaluating a summarisation tool, checking whether information has been extracted correctly or mapping a real workflow.
They don’t necessarily need access to live customer data. We can use properly controlled or simulated data.
What I want is for a student to be able to say, “This is what I built, this is the problem it solved, this is how I tested it and this is where it can fail.”
That is much more valuable than simply saying, “I completed a course.”
What skill gaps do you see in new technology professionals entering healthtech and other sectors?
From my experience, the biggest gap is not always technical knowledge.
Someone may know how to code, test software or use an AI tool, but still not understand the business process they are working on.
A good QA person in healthcare needs to understand what a referral is and what makes it complete. Someone working on a legal system needs to understand how a case actually moves through the organisation. A business analyst needs to understand what the people doing the work actually experience.
That is why I want AAIIT to combine technology with business understanding.
The future professional cannot afford to be only a coder, only a tester or only a business person. They need to understand how those disciplines connect.
How can MEDCUBE USA’s senior people contribute to AAIIT’s curriculum and mentoring?
Industry experience is extremely valuable in education because it brings reality into the classroom.
Someone who has actually dealt with a customer, a production problem, a failed release or a difficult business process can teach things that are difficult to learn from a textbook.
I would like our people to contribute through guest sessions, mentoring, case studies, project reviews and curriculum discussions.
The objective is not to turn AAIIT into a MEDCUBE USA training centre. It is to make sure that what we teach reflects what industry actually expects.
Students should understand that a system does not succeed simply because it works on a developer’s laptop. It has to work for the customer, the business and the people who use it.
What are the main hurdles when scaling AI powered solutions internationally?
One of the biggest lessons from working across markets is that you cannot assume that one solution fits everybody.
Different countries have different regulations, customers have different processes, and organisations use different systems.
The technology may be scalable, but the business process around it may not be.
That is where people and discipline become important. You need good documentation, consistent quality, trained teams and the ability to adapt to the customer’s environment.
I see a strong opportunity for Sri Lanka here. We have capable people and a growing technology sector. The challenge is to move beyond simply providing labour and start providing higher value capability. People who understand technology, business and the customer’s industry.
How do you balance fast AI commercialisation with the slower validation cycles required in regulated industries?
I make a clear distinction between an experiment and a production system.
We should be able to experiment quickly. That is how innovation happens.
But if we are going to put an AI system into a live healthcare or legal workflow, the standard has to be higher. We need proper testing, security, privacy controls, validation and human oversight.
I don’t see speed and discipline as opposites. You can move quickly in the early stages, learn from the results and then introduce stronger controls as the risk increases.
The mistake is to take a prototype and assume it is ready for production.
That is where experienced management has to step in.
Do you see opportunities for new businesses or products emerging from AAIIT and MEDCUBE USA?
Absolutely, but I don’t want to create startups simply for the sake of saying that we have startups.
The best ideas usually come from people encountering the same problem repeatedly.
If our teams or clients are spending hours dealing with a particular document process, case management problem or information bottleneck, that is where I would start looking for an opportunity.
Some ideas may become internal tools. Some may become products. Occasionally, one may become a standalone business.
AAIIT can provide the environment for people to experiment, while the MEDCUBE USA ecosystem can provide exposure to real business problems.
The important thing is to move from problem to prototype to validation, and only then think about commercialisation.
What emerging technology trend do you think will have the biggest impact on MEDCUBE USA’s roadmap?
I am watching agentic AI very closely, but I am not talking about giving an AI agent unlimited freedom.
I am interested in bounded agentic workflows.
For example, imagine a system receiving a document, identifying what it is, checking whether the required information is present, preparing the appropriate task, sending it to the right person and then tracking what happens next.
That could be very powerful in healthcare, legal case management and document workflows.
But the important word is ‘bounded’. There need to be permissions, controls, audit trails and human intervention where appropriate.
That is where I see the next stage of AI going—not simply chatbots answering questions, but AI becoming part of the workflow.
And again, that brings us back to compliant AI. The more autonomous these systems become, the more important governance becomes.
What final message would you like to share?
My message to professionals is quite simple: don’t be afraid of AI, but don’t depend on it blindly either.
Learn the fundamentals. Understand your industry. Learn how to solve problems. Learn how to work with AI and learn how to question it.
The people who will do well in the future will not necessarily be the people who know the most about one particular AI tool, but the people who can combine technology with judgement, business understanding and human skills.
While businesses will have enormous opportunities, they still need to adopt AI with discipline.

