Artificial Intelligence (AI) is at the forefront of new technology—especially in healthcare. In fact, 86% of U.S. health plan executives are eager to adopt AI, if they haven’t already.
We asked Vijay Bharadwaj, Machinify’s Chief Data Scientist, what drew him to the field of artificial intelligence, and interestingly, he ended up in AI by way of necessity.
After studying operations research in grad school, Vijay’s primary focus was improving business operations. Given the impact AI has on efficiency, his work and AI began to tightly intertwine. But, Vijay wants to make it very clear: he’s not an “AI guy.” He said, “I’m not running around with a hammer looking for nails. I’m not trying to make AI work just for the sake of AI.” Instead, he looks for business problems to solve, and AI just happens to solve a lot of them.
After years as the VP of Product Innovations, Algorithms at Netflix, Vijay brought his expertise to Machinify.
Recently, we sat down with him to ask all the important AI questions. Let’s get into it.
1. How has your approach to building AI systems evolved over the years?
Vijay: It hasn’t really. I always considered business problems to be the key. What has changed, however, is the size and scale of the problems that AI can address.
In the past, AI was used for very specific tasks, away in a corner, operating in a vacuum. Moreover, we had to very carefully control the situations where the AI system could be used.
Now, however, AI can be more involved, execute more tasks, and be more strategic in the execution of those tasks. This allows us to dream bigger in terms of what AI can be allowed to do. But the task of designing the model for how the AI system interacts with human experts, who are ultimately responsible for the decisions, has become even more important to get right.
2. What’s one project or moment in your career that fundamentally shaped how you think about AI?
Vijay: The very first problem that I worked on was for a hotel chain. I had to write code to insert an AI model to recommend prices at ~50 individual hotel locations to maximize revenue. We used to have to create a custom AI for each task.
That seems quaint compared to the scale at which AI is deployed now. For example, in the first half of 2025 alone, the AI system deployed in our audit product analyzed over 5 million medical records to extract information relevant to hundreds of different diagnoses and procedures.
Advances in AI have changed the trajectory of how work gets done but also bring up challenges of doing it carefully and safely.
3. What lessons from your time before or at Machinify have had the most significant impact on how you view AI strategy today?
Vijay: The biggest challenge for AI is the lack of data. Once you get out of a model or testing, there is a lot needed for the model to work that just isn’t available. Not only that, but approximations that exist in the real world don’t fit the academic model.
Therefore, the biggest lesson I’ve learned is to make a successful AI model, you need to have the right foundations in place.
4. There’s a lot of hype around AI—what do you think is real, and what’s still misunderstood?
Vijay: First, let’s cover what is real. Based on our experience, properly implemented AI systems can substantially reduce manual workload in healthcare operations.
However, that doesn’t mean we reduce the number of humans in healthcare. We want humans to be doing the work that is truly hard to do.
Take complex payment systems, for example. There are auditors who are trained nurses and doctors spending their time reviewing records. It takes over 10 years of study to get to that role—you don’t want talent like that searching through PDFs. Instead, you want an AI “assistant” to help extract information. Human experts will always need to oversee complex decisions.
Now, the danger of the hype—what people may misunderstand—is that implementing AI without appropriate guardrails can lead to various types of mistakes. At Machinify, through careful system design and human oversight, we’ve built our approach with this in mind.
5. How do you define “agentic AI,” and what excites you about it?
Vijay: The first stage of AI was completing really low-level tasks. For Machinify, this was data extraction for unstructured text and images. One example might be to find all the values of blood pressure measurements on a patient during a hospital stay from a PDF medical record.
The next stage is agentic AI. These advanced systems can take a high-level task or question and break it down into steps. The AI creates a structured plan, executes each step in sequence, and validates intermediate results to provide a response.
For example, an agentic AI system analyzing a medical record may be asked to determine if the patient had a hypotensive episode during the hospital stay. To answer this question, an agentic system needs to understand what a hypotensive episode means and generate a plan that involves extracting all the blood pressure measurements in the medical record, finding “normal” ranges of blood pressure values, and then checking if there was a prolonged period where the patient’s measured blood pressure was lower than normal.
Clearly, the complexity of implementing such an agentic AI system goes way beyond just simple data extraction. That’s where we’re headed.
Currently, Machinify searches medical records and pulls data, then, using another AI system, analyzes the results and compiles recommendations. By the end of the quarter, we’re working to evaluate whether or not our AI can accurately validate a correct diagnosis.
All the work will be checked by a qualified human expert to ensure accuracy. But as our AI gets progressively smarter and more accurate, the human expert will only have to verify the work done by the AI system rather than having to do all the work themselves.
6. What makes Machinify’s approach to AI different from other vendors in the healthcare space?
Vijay: It is hard to stand out with everyone calling themselves an AI company now. But, that being said, we’re different because we focus on identifying and fixing issues so that we get it right the first time and eliminate friction and inefficiency. Businesses, especially in the healthcare industry, have to be very careful about the way we apply AI.
On the data mining side, for example, we don’t let the AI determine the payment issues and say, “this is incorrect” because we must be able to explain why it’s wrong. Instead, we translate contract or regulation documents into code using AI. This creates transparency where every decision has a clear rationale tied directly to the underlying contract language or regulation.
That’s what sets us apart. We’re not just calling ourselves an AI company, we’re asking “what is the right application” while maintaining the original process. Ultimately, we’re a marriage of a tech company and deep industry expertise. Putting those two together is important for success in the healthcare world.
7. What do you think the healthcare AI landscape will look like in five years?
Vijay: The next five years are going to be about changing the processes that have existed for decades to adapt to the availability of AI models. These models can make better recommendations and automate repetitive resource-intensive tasks.
To do this, however, there needs to be careful planning.
Let’s compare AI in healthcare to the advent of electricity and its impact on the fashion industry. Sewing machines used to be powered by a massive central steam engine. When electricity was introduced, it was supposed to make everything cheaper and easier to use. However, they just replaced the steam engine with an electric motor and continued using the old pulleys and pipes to operate a web of sewing machines.
Real innovation came when smaller motors were attached to each individual sewing machine, allowing factories to completely rearrange their floor plans and, as a result, optimize how the factories were run, leading to massive efficiency gains. New technology, at face value, doesn’t immediately lead to massive productivity gains—it’s how you use it.
With AI, the next five years are going to be about rewiring existing processes. As I mentioned earlier, I expect that the job of the human experts will change from actually doing the work to just verifying the correctness of the AI systems that will do the work for them. Our goal is to enable substantial productivity improvements for human experts while increasing the accuracy of reviews.
This will allow payers to shift a significant portion of their payment integrity operations to prepay, thus reducing provider abrasion while also ensuring accurate payments for a larger fraction of the claims they receive.
8. What’s next for Machinify AI? Any sneak peeks you can share?
Vijay: Without saying too much, we’re looking to expand coverage across more healthcare processes. We’re developing capabilities to enhance recommendations and create measurable efficiencies in claims processing through our AI models. There are multiple aspects of the claims processing side of things that can be enhanced through the introduction of AI. Stay tuned.
Growth Requires a Shift in Mindset
As Vijay points out, introducing AI into the claims adjudication process is more than just plug-and-play. Swapping in AI for a singular piece of the payment integrity continuum is the old way of doing things. Sure, it’s using AI, but it’s not using AI to its full potential.
Give every sewing machine its own motor—let AI empower your teams and work for your entire PI program, not just bits and pieces.
To learn how Machinify can inject AI throughout the entire claims continuum, contact us today.
The content on this page is subject to our Terms of Use.
