Sanzhar Myrzagalym, Kuanysh Idrissov, and Zharaskhan Aman built their careers in tech companies and startups before joining forces to create Defect AI, an AI service that reviews medical documentation and helps clinics reduce the risk of financial losses caused by errors in it. This gives doctors more time to focus on patients, while clinics can lower the risk of delays and denied payments for services they have already provided.
In Kazakhstan, around 30 clinics use the solution every day. In the US, the team is developing the product under the Rette brand. It has already launched pilot projects with orthopedic clinics and is working with Mayo Clinic, one of the country’s leading medical organizations, with hospitals, clinics, and research centers under its umbrella.
In an interview for Digital Business and Astana Hub’s joint project, “100 Startup Stories from Central Eurasia”, Sanzhar and Kuanysh spoke about how the team entered the US market, why it chose to focus on orthopedics, and why clinics need Rette.
“For the first time, there’s a real opportunity to automate medical documentation”
– Tell us about yourselves. What were you doing before launching the startup?
Sanzhar: – Kuanysh and I have known each other since our first year at KBTU, so for more than 15 years now. Over that time, we’ve worked on a lot of different projects together. One of them was Sheberkhana, a prototyping lab based at Astana Hub, which we launched in 2018.
After earning my bachelor’s degree in Computer Science, I spent nine years at Microsoft, where I worked as a data and AI architect, designing AI solutions for large enterprise clients. More recently, I completed my master’s degree at Stanford.
Sanzhar Myrzagalym
Kuanysh: – I’ve been leading product teams for more than 10 years. For the three years before starting my own company, I was a product lead at AlemHealth, a Singapore-based HealthTech startup that went through Y Combinator. The company built solutions for radiology and remote diagnostics, used by thousands of doctors across more than 40 countries.
Kuanysh Idrissov
– How did you come up with the idea for your own business?
Sanzhar: – Kuanysh and I had been working on different projects together since our student days, and we always wanted them to be genuinely useful to someone. Over time, we realized that if we wanted to make a bigger impact, we had to commit to it full-time rather than only working on projects in our spare time.
Two years ago, I decided to leave Microsoft. My career there was going well, but I knew I needed time to figure out where I could make the biggest difference. A master’s degree felt like a good way to take that pause and rethink what I wanted to do next. I decided that if I was going to continue studying, it had to be at one of the world’s top universities. Stanford was first on my list, and I got in and moved to the US.
Even while I was at Microsoft, I could see how quickly AI was developing, and I believed healthcare was one of the areas where it could have the greatest impact. At Stanford, I studied systems thinking and decided to apply it to healthcare. I spent the past year studying the US healthcare system, practically living at Stanford School of Medicine, trying to understand how it works from the inside and what could be improved.
Kuanysh: – At AlemHealth, I worked with hundreds of doctors from different countries and saw the problems they ran into most often. Sanzhar and I were constantly comparing notes on what each of us was learning from our own work. Over time, those conversations led us to medical documentation.
A lot depends on it. First, the quality of care itself: anyone who works with a patient later needs a complete and accurate record. Second, the financial side: in many countries, clinics are paid through public or private insurance systems, and if the documentation contains errors or is missing required information, the clinic risks not being paid for care it has already provided.
Sanzhar: – I think the main reason we decided to build a healthcare startup was a unique combination of circumstances. Kuanysh had firsthand insight into the problems doctors were facing, while I was studying the healthcare system at Stanford. At the same time, we were both closely following advances in AI and realized that for the first time, there was a real opportunity to automate complex documentation-related processes in healthcare.
And the market is enormous. In the US alone, around $5 trillion is spenton healthcare every year, equivalent to 18% of the country’s GDP.
Once we saw the opportunity, we decided to focus on it completely. When we started working on the project, we invited Zharaskhan Aman to join the team. He’s also a KBTU graduate. He took part in programming competitions and twice reached the finals of the prestigious ICPC. He later worked as an engineer at Meta, followed by roles at several startups.
For the first few months, we funded the work ourselves. Then we secured backing from investors, and the project became Defect AI.
“Document reviews used to take days. Now they take about five minutes”
– What problem does Defect AI solve?
Sanzhar: – In short, we help clinics keep their medical documentation complete, accurate, and free of gaps. Everyone depends on those records: patients, doctors, and the insurance system that pays for care. But keeping them up to date takes a huge amount of time.
In the US, there’s even a term for it: pajama time, meaning the hours doctors spend at home in the evening finishing documentation after seeing patients. The situation in Kazakhstan is similar, there just isn’t a name for it. When medical records are done properly, doctors and other staff have more time for patients, while the organization reduces the risk of payment delays or denials.
A significant share of healthcare services in Kazakhstan is paid for through the mandatory health insurance system. The process is straightforward: a patient receives treatment, and the healthcare provider then submits the documentation to the health insurance system for payment. There are a lot of requirements, and getting everything right the first time isn’t easy. We help check whether the care provided has been documented fully and accurately, and flag anything that should be reviewed before submission.
Kuanysh: – The requirements are also updated regularly, whether through changes in legislation, clinical protocols, or official regulations. Keeping track of all of them is a major task in itself for clinic staff. Defect AI takes that work on. Before documents are submitted, it checks them against the latest requirements and shows what information is missing and what needs to be clarified. We structure the entire data flow and point to the specific areas that need attention, so clinic staff can focus only on those issues.
– What results have you achieved?
Sanzhar: – Our solution is now used every day by around 30 clinics in Kazakhstan, and we’re adding new ones every week. The biggest thing we see is the change over time: in the first few days, the system usually finds quite a few places where information is missing from the records. A few months later, the number of those issues drops several times over.
It also saves staff a lot of time. Reviewing documents before submission used to take days. Now it takes about five minutes.
Kuanysh: – After a few months, we knew the product was working, but we also saw that the Kazakhstani market was relatively small, with around 2,000 healthcare organizations in the insurance system. So we started looking for ways to scale beyond Kazakhstan.
But Kazakhstan gave us the most important thing: proof that the approach works. We also received significant support from Astana Hub along the way. We went through acceleration programs, took part in Hub events, and the tax incentives available at an early stage helped reduce the administrative burden and allowed us to direct more resources toward the product and the team.
Now we’re following the same path in the US, just in a market that’s hundreds of times larger.
“Rette comes from the Kazakh word retteu, meaning to put things in order”
– How did you choose which market to enter next?
Sanzhar: – We did our research. Naturally, we considered neighboring countries. We also looked at Southeast Asia: in Indonesia and Malaysia, technology adoption is at roughly the same level as in Kazakhstan, and the healthcare systems are fairly similar. Europe was another option, but despite the EU framework, each country has its own regulatory requirements, which creates significant limitations.
In the end, we decided that if we were going to enter a new market, we might as well go for the biggest one. And that meant the US.
– How much did you have to adapt the product?
Kuanysh: – The US system is much more complex. Of the roughly $5 trillion spent on healthcare every year, around $1 trillion goes toward administrative costs.
There are thousands of insurance companies, each with a huge number of plans and its own set of rules. We saw an opportunity to help clinics bring some order to all of this and build a large business around it. Even the name reflects that: Rette comes from the Kazakh word retteu, meaning “to put things in order.” In the US, the name Defect AI sounded too negative to people.
Rette is already a broader product. It works with documentation at every stage of the patient journey, from getting treatment pre-authorized by the insurance company to submitting a claim for payment and filing an appeal if that claim is denied. We started with orthopedics.
– Why did you decide to focus on one specialty instead of building a product for the entire market, especially since its size was one of the reasons you chose it?
Sanzhar: – A lot of companies take a broad approach and try to build one big product that can meet the needs of any clinic. We chose a different path: start with one specialty and really understand its specific challenges.
Every specialty is very different. One expert put it this way: US healthcare may be a $5 trillion market, but in reality, it’s made up of thousands of billion-dollar markets. You can’t boil the ocean. You have to start with something very specific, solve the problems of one group of doctors, and expand from there.
Orthopedics involves a lot of expensive procedures, some costing up to $250,000, as well as complex approval processes with insurers. That makes the cost of documentation errors particularly high.
– Let’s take an example. A patient comes to see a doctor at a US clinic. What happens next, and where does Rette fit into the process?
Sanzhar: – The doctor sees the patient and records the relevant information in the clinic’s medical information system. That clinical note has a major impact on how the rest of the process unfolds.
If the patient needs surgery, for example, the clinic will often have to go through prior authorization first and get approval from the insurance company. A lot depends on how thoroughly the doctor has documented the case. If a required indication is missing from the note, the insurer may deny the procedure.
Once the approval is in place, the doctor provides the treatment. That generates another set of documents, which then goes to the billing team. They submit a claim to the insurer for payment, and that claim also needs to be checked against the relevant rules.
The insurer then either pays for the service or denies the claim. If it is denied, the clinic can provide additional information and file an appeal. Rette supports the clinic throughout each of these stages.
– How exactly does Rette check the data?
Kuanysh: – We have our own proprietary models, combining traditional machine learning with large language models. They check whether the records are complete and accurate. We train and validate them on real clinical data together with one of the leading hospitals in the US.
We’ve also built an agent-based infrastructure that keeps the rules database up to date. Hundreds of AI agents check for updates every day on insurance company websites and portals. Tracking that volume of changes manually would be almost impossible.
– If the system finds an issue, what does the doctor or clinic staff member see?
Sanzhar: – The doctor creates the medical record as usual. Our system then checks it and, if it spots any risks, sends an alert. For example, it may flag that certain information is missing and that the record could run into problems later in the process.
Right now in the US, this check happens after the document has already been created. But we’re moving toward real-time review, where the doctor writes the note and the system immediately shows what information still needs to be added.
“A clinic provides $1 million worth of care, then spends another $100,000 just to get paid that $1 million”
– How much time and money does Rette save a clinic?
Kuanysh: – By our estimates, at some healthcare organizations, the cost of getting paid for services already provided can reach 10% of revenue. In simple terms, a clinic delivers $1 million worth of care, then has to spend another $100,000 just to collect that $1 million. That pushes the cost of already expensive healthcare even higher.
Our goal is to bring that figure down to around 1% by automating much of the process and helping clinics collect the same amount of revenue at a much lower cost.
– How do you make money from the product?
Sanzhar: – In Kazakhstan, we charge a subscription fee. In the US, we plan to combine a subscription with a percentage of the revenue we help clinics recover. At the moment, we expect that to be around 1%, which is significantly lower than what the market typically charges for this kind of work.
So, for example, if a clinic provides $1 million worth of care, we would earn $10,000 for helping make sure the documentation is in order.
– What have you achieved in the US so far?
Kuanysh: – We launched our first pilot projects with orthopedic clinics in the spring. We’re now wrapping them up and expect them to turn into commercial contracts soon. One of the pilots is with one of the largest independent orthopedic groups in California, and we’re also in talks with clinics in several other states.
Rette was selected for the Mayo Clinic Platform Accelerate program, run by Mayo Clinic, one of the leading medical organizations in the US with a network of hospitals and clinics. As part of the program, we train and validate our models on real clinical data and work with advisors from across the Mayo Clinic network. We’ve also signed an agreement with the California Orthopaedic Association. Together, we’re working to incorporate the association’s clinicalguidelines into our product.
Another important part of the story is StartX, an accelerator in the Stanford ecosystem that we joined with support from Astana Business Campus. It opened a lot of doors for us. Through the StartX community, we found several strong advisors, secured angel investment, and were introduced directly to a number of clinics.
TechCrunch Disrupt 2025, San Francisco
– How are you funding the project?
Sanzhar: – For the first six to eight months, we funded the project from our own savings.
Last year, we closed our first investment round with participation from 500 Eurasia, JAS Ventures, and Astana Hub Ventures. We recently opened a pre-seed round and are primarily targeting US-based funds. Our existing investors have also confirmed that they will participate.
“85–90% of our focus is now on growing the product in the US”
– Who is working on Rette now?
Kuanysh: – Defect AI and Rette share the same team, and it’s distributed across Kazakhstan, the US, Europe, and Asia. The team includes developers, a data science team, a product manager, and a commercial director responsible for Kazakhstan. We also have a lawyer who helped us set up the structure so that the Kazakhstani company and our US C-Corp operate as one business.
We’re actively expanding our team of ML research engineers. We’re building our own state-of-the-art models for medical documentation, meaning models that perform at the level of the best solutions available today, and this is one of our key advantages. General-purpose language models can read text. Ours understand clinical notes: they’re trained on real data and have detailed knowledge of orthopedics and insurance requirements. We also plan to publish some of this work, because we want to do more than build a product. We want to contribute to moving the industry forward.
We also hired a dedicated compliance specialist. That’s essential when you work with medical data. We’ve obtained ISO/IEC 27001:2022 certification, meet HIPAA requirements, and completed a SOC 2 audit. For clinics and hospitals, this gives them confidence that they can trust us with patient data.
Sanzhar: – We also work with medical experts in Kazakhstan on a part-time basis. They advise us on regulatory changes, new requirements, and official orders.
– Do the advisors work with you for free?
Sanzhar: – We have formal arrangements in place and have signed advisory agreements. Advisors are usually compensated with equity in the startup. But more importantly, these are people who genuinely believe in what we’re building. Our advisors include the head of an orthopedic department at one of Mayo Clinic’s hospitals, as well as practicing surgeons. We also work with entrepreneurs who have built HealthTech companies themselves and were in a very similar position to us just a few years ago.
– What do you plan to achieve by the end of 2026? And looking further ahead, what do you ultimately want Rette to become?
Kuanysh: – 85–90% of our time and attention now goes into growing the product in the US. Our plans for 2026 are to turn the current pilots into commercial contracts and increase their number. In California, we want to establish ourselves with the state’s largest independent orthopedic group and, together with the California Orthopaedic Association, roll out our documentation standard across all of its clinics. At the same time, we want to move discussions in other states into the pilot stage. With Mayo Clinic, the plan is to keep building the partnership, start working with clinics across its network, and publish our first joint research papers.
At the same time, the service continues to grow in Kazakhstan. When clients here ask for something new, we can usually add it to the product within just a few days, and we’re bringing new clinics on every week. But our bigger goal is to help improve healthcare in Kazakhstan, raise the quality of care, and reduce the workload on doctors.
Sanzhar: – That’s why, alongside our work with clinics, we also collaborate with educational institutions, including Astana Medical University and Nazarbayev University School of Medicine. We want to improve the quality of healthcare in Kazakhstan at a systemic level, not just through one product. If future doctors learn to handle medical documentation properly and use modern tools while they’re still studying, they can carry those skills into clinical practice later on.
Looking at the bigger picture, we want Rette to become a trusted layer between clinics and insurers. Ideally, if a medical record has gone through Rette, it should be treated as complete and accurate. The insurer would not need to spend resources checking it all over again. It could trust the data and move to payment faster.
That would leave doctors to focus on treating patients instead of paperwork, help clinics get paid what they have earned without unnecessary losses, and allow patients to receive treatment sooner. It may not happen next year. It may take much longer. But that is what we started all of this for.