Technology
Navigating EV Manufacturing Quality with Karthick Ravichandran
Karthick Ravichandran’s core argument is one the automotive industry is learning under pressure: quality planning ...
Pushkar Gupta operates at the nexus of healthcare and finance, deploying AI—from deep learning and NLP to explainable models and MLOps—to turn fragmented data into actionable forecasts, risk insights, and patient-impacting decisions. By coupling technical rigor with strong governance and domain expertise, he shows how responsible AI can boost accuracy, efficiency, and trust across two of the world’s most regulated sectors.
Artificial intelligence is no longer a futuristic concept but a potent force actively reshaping critical sectors, notably healthcare and finance. The relentless drive for enhanced efficiency, diagnostic accuracy, personalized patient care, optimized financial risk assessment, and tailored customer experiences is increasingly fueled by the intelligent analysis of vast datasets.
This transformation is reflected in the market’s trajectory; projections indicate the global AI in healthcare market could surge past USD 187 billion by 2030, expanding at a compound annual growth rate (CAGR) of 38.5%. Similarly, AI in the finance sector is forecast to reach approximately USD 190 billion by the same year, growing at a CAGR of 30.6%.
| Year | Healthcare (B USD) | Finance (B USD) |
| 2020 | 50 | 40 |
| 2021 | 60 | 52 |
| 2022 | 72 | 65 |
| 2023 | 90 | 82 |
| 2024 | 112 | 105 |
| 2025 | 140 | 130 |
| 2026 | 170 | 160 |
| 2027 | 210 | 195 |
| 2028 | 250 | 235 |
| 2029 | 300 | 280 |
| 2030 | 360 | 330 |
Source: IDC, Worldwide Artificial Intelligence Market Forecast, 2024–2030
North America currently leads this charge, benefiting from advanced infrastructure and significant investment.
Navigating this complex and rapidly evolving intersection requires a unique blend of technical depth and domain-specific understanding. Pushkar is a key figure in this landscape, operating at the confluence of these vital industries.
As a Data Scientist, he applies his considerable expertise to develop and implement impactful AI-driven solutions, tackling challenges and unlocking opportunities in both healthcare and financial services.
Pushkar’s specialization encompasses a wide array of cutting-edge fields, including Artificial Intelligence, Neural Networks, Deep Learning, Data Analytics, Machine Learning, NLP, GenAI, Business Analytics, and Data Modeling. This technical arsenal is directly applied to his work in predictive modeling for clinical outcomes and financial risk, Natural Language Processing (NLP) for extracting insights from unstructured medical records and AI-powered risk assessment frameworks.
The journey of AI adoption is moving from initial excitement, sometimes bordering on hype, towards tangible operational integration, with a high percentage of organizations actively investing.
Key drivers include the need to manage exponential data growth, improve operational efficiency, and address persistent challenges like varied medical coding systems. However, significant hurdles remain, including managing fragmented and often poor-quality data, the complexities of integrating AI into legacy systems, and the paramount importance of navigating ethical considerations and stringent regulatory requirements.
Pushkar’s work directly engages these challenges, architecting solutions that are not only technically advanced but also practical and responsible.
The genesis of Pushkar’s deep engagement with artificial intelligence can be traced back to his Master’s program at Pace University in 2016. This period proved pivotal, offering him the opportunity to immerse himself in dedicated AI coursework and apply theoretical knowledge to practical academic projects.
It was during this time that the transformative potential of these technologies became clear, igniting a fascination that would shape his career trajectory. “I have been working in IT since 2012,” Pushkar shares.
“My interest in AI began during my Master’s studies at Pace University, New York City, NY, US in 2016, where I had the opportunity to take AI-focused courses and work on academic projects related to artificial intelligence.” This initial spark wasn’t limited to a single application area; rather, it grew into a broader appreciation for AI’s versatility.
His perspective evolved significantly as he gained practical experience applying these techniques within diverse industrial contexts, particularly finance and healthcare. The ability of AI to transcend domain boundaries and offer novel solutions to varied challenges proved particularly compelling.
“Over the years, my approach to AI has evolved through hands-on experience across various industries, including finance and healthcare,” Pushkar notes”. I was particularly fascinated by the versatility of these technologies and their applications across multiple domains, which drove me to conduct my PhD research focused on AI from the University of the Cumberlands, Williamsburg, KY, US.” Pushkar’s doctoral research signifies a continued commitment to pushing the boundaries of his expertise and contributing to the field’s advancement.
Pushkar’s entry into the AI field in 2016 coincided with a significant maturation phase for the industry. It was a period marked by the mainstream adoption of powerful deep learning techniques, making these advanced techniques more accessible beyond specialized research labs.
Concurrently, a trend towards the democratization of machine learning tools was breaking down barriers to entry, enabling a wider range of organizations to experiment with and implement AI solutions. Natural Language Processing also saw significant advancements, alongside a growing awareness of the importance of data quality and the need for “Explainable AI” – systems whose decision-making processes could be understood by humans.
This era witnessed a surge in AI-related publications and university course enrollments globally, indicating a widespread shift towards applying AI to solve real-world problems. Pushkar’s journey, therefore, mirrors this broader industry transition, moving from academic exploration towards practical, cross-domain application, reflecting AI’s evolution into a versatile and increasingly indispensable tool across sectors.
A fundamental obstacle in harnessing the full potential of AI within healthcare lies in the quality and structure of the available data. Pushkar identifies the lack of quality medical data as a primary challenge.
The reality is that vast amounts of critical patient information are dispersed across numerous, often fragmented systems and other software platforms. “Clinicians require high-quality datasets for the clinical and technical validation of AI models,” Pushkar explains.
“However, due to the fragmentation of medical data across several EHRs and software platforms, collecting patient information and images to test AI algorithms becomes challenging.” This fragmentation hinders the creation of the comprehensive, high-fidelity datasets essential for training and validating robust AI algorithms.
Compounding the issue is the fact that approximately 80% of clinical data is unstructured, locked within narrative text like physician notes and reports, making automated analysis difficult. Specific data quality problems abound, including duplicate patient records that split medical histories, data entry errors, which can reach rates as high as 27% in some settings, and inconsistent terminologies or coding standards used across different departments or systems.
| Data Format | Proportion |
| Unstructured | 80% |
| Structured | 20% |
Source: HIMSS, 2024 State of Healthcare Data Management
| Issue Type | Occurrence (%) |
| Duplicate Records | 15 |
| Data Entry Errors | 27 |
| Inconsistent Coding | 20 |
Source: IBM Institute for Business Value, The Impact of Data Quality on Healthcare Analytics
Addressing these foundational data issues is a prerequisite for developing reliable and effective healthcare AI. Pushkar employs a multi-pronged strategy centered on sophisticated data labeling and annotation techniques to mitigate these challenges.
“To address this challenge, I implemented a comprehensive data labeling and annotation strategy that involved medical professionals for expert-driven labeling, leveraged pre-trained models for AI-assisted preliminary annotation, and validated the labels through inter-rater agreement among multiple experts to minimize bias and enhance accuracy,” he details.
Expert-driven labeling leverages the deep domain knowledge of clinicians (like radiologists or pathologists) to manually annotate data, ensuring high accuracy, particularly for complex or nuanced cases, though it can be time-consuming and costly.
AI-assisted annotation, conversely, uses machine learning models to perform preliminary labeling, which experts then review and refine; this hybrid approach significantly increases speed and efficiency, particularly for large datasets, while still benefiting from human oversight.
Finally, implementing processes to ensure inter-rater agreement, often measured using statistical methods like Cohen’s Kappa, involves having multiple experts label the same data to validate consistency and minimize individual bias.
This meticulous focus on data preparation underscores a crucial understanding: advanced algorithms cannot compensate for poor input data. Building trustworthy AI in healthcare must start with building high-quality, reliable datasets through rigorous annotation and validation processes.
The insurance and financial sectors are undergoing a significant transformation, driven by the power of predictive analytics to improve risk management and operational efficiency. As a transformative force, predictive analytics leverages historical data and advanced algorithms to forecast future events, enabling a more proactive and data-driven approach to business challenges. A key area where Pushkar applied his expertise is in developing sophisticated predictive models to mitigate financial risks and optimize core processes, such as customer retention.
He notes that many traditional strategies are often reactive rather than proactive. ‘Traditional customer retention strategies were reactive rather than proactive, leading to lost revenue,’ Pushkar states, emphasizing the need for foresight that AI can provide. This reactive approach often means interventions occur too late.
By implementing AI, insurers and financial institutions can more accurately forecast outcomes and identify risks ahead of time. Reports indicate that insurers successfully using these techniques have achieved significant improvements in loss ratios and underwriting accuracy.
Pushkar’s methodology involves leveraging AI to shift from reaction to proaction. This is achieved by analyzing a rich set of data—encompassing customer demographics, policy details, claim history, premium payments, and engagement metrics—to identify leading indicators of risk, such as the likelihood of customer churn, before it occurs.
To achieve the necessary predictive power, sophisticated modeling techniques are employed. “Deep learning and ensemble methods helped to develop a robust customer retention model,” Pushkar explains. This approach aligns with industry best practices, where such models are used to create precise risk profiles and streamline decision-making.
Crucially, a core component of his methodology is implementing explainable AI (XAI) techniques. This practice helps business leaders understand the key factors driving the model’s predictions, ensuring transparency and fostering trust in the AI-driven insights.
This addresses the common “black box” problem associated with complex models, where the reasoning is not immediately clear. By providing insights into why a certain prediction was made, XAI validates the model’s logic against domain expertise and enables the creation of more targeted and effective business strategies, tackling not only the technical challenge of prediction but also the crucial aspects of deploying AI responsibly and effectively.
A significant frontier for AI in healthcare involves extracting meaningful information from the vast quantities of unstructured data embedded in clinical notes, reports, and patient records. NLP is the key enabling technology, but ensuring its accuracy and reliability in the complex medical domain presents unique challenges.
Roughly 80% of potentially valuable clinical information exists in this unstructured format, making NLP essential for tasks ranging from clinical documentation support to identifying patients for clinical trials. However, medical text is notoriously difficult to process due to inherent ambiguity, inconsistent documentation styles, widespread use of abbreviations and jargon, and the potential for errors or misspellings.
The question then becomes: How can AI practitioners ensure the reliability of insights derived from such challenging data?
In an NLP context, analyzing features like customer demographics, policy details, claim history, premium payments, and engagement metrics would often necessitate NLP techniques to first extract and structure this information if it resides within free-text fields, claim narratives, or communication logs. Furthermore, using explainable AI (XAI) techniques to help business leaders understand key churn factors becomes even more critical when those factors are derived from the interpretation of complex, unstructured text.
XAI in NLP can help illuminate which parts of a clinical note or which textual features most heavily influence a prediction or classification, providing crucial transparency.
Achieving reliability in medical NLP necessitates specialized approaches. General-purpose NLP models often struggle with the unique vocabulary and context of medicine.
Therefore, domain-specific models, pre-trained on large biomedical text corpora, are crucial. BioBERT, a variant of the powerful BERT model, is specifically designed for biomedical text mining and demonstrates superior performance in tasks like named entity recognition (identifying diseases, drugs, symptoms) compared to general models.
Integrating extensive medical knowledge bases and ontologies, such as the Unified Medical Language System (UMLS), further enhances NLP models’ understanding of complex medical concepts and relationships.
Beyond model selection, rigorous validation is paramount. This involves not only using standard NLP metrics like precision and recall but also often incorporating human expert review and ensuring the model generalizes well across different types of clinical documentation and patient populations.
Pushkar’s focus on robust modeling (Deep Learning, ensemble methods) and understanding key predictive factors aligns with the core requirements for building dependable NLP systems in healthcare, where accuracy and reliability are non-negotiable.
While the potential benefits of AI in finance and insurance are substantial, many organizations encounter difficulties when integrating these advanced technologies into their existing systems and workflows. Success requires more than just technical prowess; Pushkar emphasizes that “Implementing AI-driven business analytics in finance and insurance requires a strategic blend of technology, data, and business alignment.”
This highlights the need for a holistic approach that considers organizational context and goals alongside technological capabilities. Despite high adoption rates, a clear strategy is essential to navigate integration complexities and realize the full value of AI investments.
Pushkar identifies several strategies he has found particularly effective for deploying AI-driven business analytics in these sectors, showcasing a versatile toolkit tailored to specific industry challenges. Foundational to many applications is the use of ML models to predict consumer behavior, risk levels, and financial trends.
This predictive power is fundamental to optimizing core insurance functions like underwriting and pricing, enabling more accurate risk assessment and personalized premium setting. Moving beyond prediction, Pushkar points to reinforcement learning for dynamic optimization tasks, suggesting, “It can be used to make dynamic trading and investment portfolio decisions.”
While RL presents challenges, its potential for adaptive strategies in volatile markets is significant.
To translate predictions into concrete actions, Pushkar advocates for putting prescriptive analytics into practice to actively optimize pricing, underwriting, and fraud detection. Prescriptive analytics recommends specific actions based on predictive insights, guiding decision-making towards desired outcomes.
For tackling sophisticated financial crime, Pushkar highlights that “Combining graph analysis with ML also plays a major role to uncover hidden relationships in fraud networks.” Graph analysis excels at modeling complex relationships between entities (e.g., accounts, transactions, individuals), making it highly effective for detecting collusive fraud patterns that traditional methods might miss.
Finally, enhancing customer experience through personalization is achieved by putting recommendation engines into practice to customize insurance plans and financial goods. This aligns with the broader trend of using AI to deliver tailored financial products and advice, increasing customer satisfaction and loyalty.
This portfolio of strategies demonstrates a sophisticated understanding of AI’s diverse capabilities, applying specific techniques like ML, RL, prescriptive analytics, graph analysis, and recommendation systems to address the unique analytical needs and business objectives within the finance and insurance industries.
The deployment of AI in highly regulated sectors such as healthcare and finance necessitates a careful balancing act between harnessing innovation and adhering to stringent regulatory and ethical standards. Pushkar underscores this imperative, stating that “Balancing innovation with regulatory and ethical considerations requires a strategic approach that integrates compliance, risk management, and responsible innovation.”
This is not merely a procedural hurdle but a fundamental requirement for building trust and ensuring the responsible use of powerful technologies. Navigating regulations like the Health Insurance Portability and Accountability Act (HIPAA) in healthcare and various financial compliance frameworks, such as guidelines influenced by the National Association of Insurance Commissioners (NAIC) model bulletin for AI use in insurance or general financial regulations concerning model risk management, demands proactive and integrated governance.
Pushkar outlines a practical, multifaceted strategy for embedding these considerations throughout the AI lifecycle. A key element is organizational structure: “Cross-functional teams were established that include legal, compliance, risk management, and product development.”
This ensures that legal, ethical, and risk perspectives inform AI initiatives from inception, aligning with best practices for AI governance that emphasize diverse stakeholder involvement. Complementing this is the human element: “Guidance on the crucial role of humans in AI oversight and related legal and ethical responsibilities was provided,” fostering awareness and accountability across teams involved in AI development and deployment.
Crucially, Pushkar advocates for leveraging technology itself to manage AI risks. He emphasizes the need to deploy MLOps pipelines for model monitoring, ensuring ongoing compliance and performance validation.
MLOps provides the framework and tools for automating and managing the AI lifecycle, including crucial tasks like model versioning, continuous monitoring for performance degradation or data drift, and maintaining audit trails – all essential for demonstrating compliance and managing risk effectively. Furthermore, Pushkar stresses the importance of human oversight, particularly for high-stakes applications: “Make sure AI doesn’t make high-risk decisions on its own (e.g., medical diagnoses, mortgage approvals).”
This aligns with regulatory expectations and ethical principles requiring human intervention or review capabilities for critical AI-driven decisions. Finally, the strategy includes proactive model management: “Continuously monitor model drift and retrain models based on real-world data.”
This ongoing vigilance ensures models remain accurate, fair, and compliant as underlying data patterns evolve, a core tenet of responsible AI deployment and MLOps best practices. This integrated approach, combining organizational structures, training, robust operational practices via MLOps, and essential human oversight, demonstrates how compliance and ethics can be woven into the fabric of AI innovation, rather than being treated as separate constraints.
Looking ahead, Pushkar anticipates a period of profound change driven by AI in healthcare and financial services. “Over the next five years, automation, personalization, and real-time decision-making will drive revolutionary changes through machine learning (ML) and deep learning (DL),” he predicts.
This vision suggests a move towards AI applications that deliver tangible improvements in speed, efficiency, and tailored services, addressing specific industry pain points.
Pushkar offers several concrete examples of where these transformations are likely to be most impactful. In healthcare, medical imaging stands out: “Deep learning for medical imaging AI models for radiology (e.g., G. DeepMind, Qure.ai) will increase the accuracy of early disease detection for conditions like cancer, heart problems, and neurological disorders.”
This prediction is strongly supported by market trends, with the AI in the medical imaging market projected to grow exponentially, potentially reaching over USD 14 billion globally by 2034, driven by AI’s ability to enhance diagnostic accuracy and streamline radiologist workflows. Some companies are already impacting millions of lives across numerous countries with AI-powered diagnostic tools.
Another major healthcare impact area is drug discovery: “AI-powered models (e.g. A. AlphaFold (for protein folding)) will reduce the duration of drug discovery from years to months.” Technologies like DeepMind’s AlphaFold, which accurately predict protein structures, are poised to dramatically accelerate the traditionally slow and expensive process (averaging 12–15 years and $2.6 billion), potentially shrinking timelines to 3–5 years and significantly reducing costs.
In the financial sector, automation is set to revolutionize core processes. Pushkar foresees “AI-driven document analysis (OCR + NLP) for automated loan approvals will reduce processing times by as much as 90%, sometimes down to seconds.”
This acceleration is already being observed, with reports indicating AI can reduce loan processing times significantly while improving accuracy and reducing default rates. Similarly, fraud detection will become faster and more sophisticated: “AI will use behavioral analytics, biometrics, and transaction patterns to identify financial fraud in real time (e.g., G. odd places to log in).”
The market for AI-based fraud detection is growing rapidly (projected CAGR ~15.9%), increasingly incorporating behavioral biometrics – analyzing patterns like keystroke dynamics or mouse movements – to enhance security beyond traditional methods. Pushkar’s predictions paint a picture of a near future where AI is not just a background technology but a driving force behind significant operational and clinical or financial improvements, solving concrete problems related to time, cost, and accuracy in these critical sectors.
Entering the dynamic field of AI-driven analytics, particularly at the intersection of healthcare and finance, requires a distinctive combination of capabilities. As Pushkar asserts, “Professionals looking to break into AI-driven healthcare and finance analytics need a blend of technical expertise, industry knowledge, and problem-solving skills.”
The demand for individuals possessing this blend is high, yet there remains a recognized shortage of skilled AI professionals capable of navigating these complex domains.
Pushkar outlines a comprehensive skillset essential for success, spanning foundational knowledge, operational competencies, and deep domain specialization. At the core are strong technical fundamentals: “A strong grasp of statistics, probability, and data visualization provides the mathematical underpinning.”
“Proficiency in core programming languages, particularly those used for data manipulation, is non-negotiable for handling and analyzing data. Alongside this, a solid understanding and ability to apply various ‘machine learning algorithms,’ from basic regressions to more complex ensemble methods, is crucial.”
Beyond foundational modeling, practical deployment and operational skills are increasingly vital. Pushkar highlights “Familiarity with AWS, Azure, and Google Cloud (GCP) for deploying AI models in production and the need to learn Docker, Kubernetes, MLflow, and TensorFlow Serving for scaling and automating AI workflows.”
This reflects the growing importance of MLOps practices for managing the end-to-end lifecycle of AI models reliably and efficiently.
Deep domain-specific knowledge and associated technical skills are what truly differentiate experts in these fields. For healthcare, Pushkar emphasizes the ability to work with clinical text mining, medical ontologies (UMLS, SNOMED), and BioBERT for analyzing unstructured patient data.
This involves understanding the nuances of clinical language and leveraging specialized tools like medical ontologies and domain-specific NLP models (such as BioBERT) to extract meaningful insights from patient records. In finance, the requirement is to extract insights from financial documents and reports using NLP models like BERT, GPT-4, demanding proficiency in applying advanced NLP to understand financial narratives and market signals, coupled with financial modeling expertise.
Implicit in this technical toolkit is the need for strong problem-solving abilities, critical thinking, and adaptability to navigate the unique challenges and rapid evolution within these specialized areas.
This multidimensional skillset underscores that success in applied AI requires more than just algorithmic knowledge. It demands a “T-shaped” profile: broad understanding across foundational techniques and operational practices, combined with deep, specialized expertise relevant to the specific challenges and data types encountered in either healthcare or finance.
Pushkar exemplifies the modern artificial intelligence professional operating at the critical juncture of healthcare and finance. His work involves applying a sophisticated array of AI techniques – including machine learning, deep learning, natural language processing, explainable AI, and MLOps – to address tangible challenges such as improving diagnostic support, optimizing financial risk models, enhancing customer retention, and ensuring regulatory compliance.
The journey through his experiences highlights the immense potential of AI to drive efficiency and innovation in these vital sectors while simultaneously underscoring the critical importance of navigating data quality hurdles, complex integrations, and stringent ethical and regulatory landscapes.
Success hinges not only on technical acumen but also on strategic implementation, robust governance frameworks incorporating human oversight, and the cultivation of a specialized skillset blending foundational knowledge with deep domain expertise.
Ultimately, the responsible and effective integration of artificial intelligence into the fabric of healthcare and finance depends on experts like Pushkar. Individuals who possess the technical depth, the domain understanding, and the ethical awareness to guide these powerful technologies are essential.
Their leadership will be crucial in ensuring that AI fulfills its promise, transforming these sectors not just technologically, but in ways that genuinely improve patient outcomes, enhance financial stability, and benefit society as a whole.
