Can AI help healthcare systems make sense of their data?
As healthcare systems continue to amass vast amounts of patient information, the question arises: how can this data be effectively utilized to improve patient outcomes? Gaurav Tripathi, Co-founder and Group CTO at Partex.AI, presented at DevSparks 2026 in Pune, shedding light on innovative technologies such as federated learning, knowledge graphs, and agentic AI. These technologies promise to tackle the challenge of fragmented medical data while safeguarding patient privacy, suggesting a future where data-driven decision-making is both effective and ethical.
The Fragmented Nature of Healthcare Data
Healthcare organizations utilize multiple digital systems—ranging from electronic health records (EHRs) to laboratory reporting tools—to manage patient information. Although these systems capture critical medical information, they operate independently, leading to fragmented datasets. This fragmentation complicates the identification of patterns within patient histories and population health trends.
Furthermore, advances in medical research increasingly rely on robust data analytics. Detecting patterns across diagnostic reports and treatment outcomes can enable clinicians to diagnose diseases earlier and design more effective therapies. However, sharing sensitive medical data across institutions introduces significant privacy concerns, as healthcare data is among the most personal and sensitive categories of information.
Data Privacy is Non-Negotiable
Data privacy in healthcare is paramount. During his presentation, Tripathi emphasized that safeguarding patient records is non-negotiable, especially when leveraging vast datasets for research and analytics. This perspective informs the way AI technologies—specifically federated learning and knowledge graphs—should be developed and implemented in healthcare.
Sovereignty Beyond Data Localization
The discourse on digital sovereignty often zeroes in on where data is stored. Many countries champion keeping sensitive data within national borders to maintain regulatory oversight. However, Tripathi raised an important point: sovereignty in healthcare extends beyond data localization. Healthcare systems differ widely across regions, and AI models trained primarily on global datasets may not accurately reflect local healthcare realities.
For instance, India's healthcare infrastructure, demographics, and disease patterns diverge significantly from those in Europe or the United States. Models tuned to international datasets may neglect critical nuances of the Indian healthcare system. Therefore, Tripathi posits that sovereign healthcare models must adapt to local needs, respect multilingual environments, and adhere to regulatory frameworks like India’s Digital Personal Data Protection law.
Training AI Models Without Moving the Data
One promising technology for tackling these challenges is federated learning. Traditional AI frameworks typically require data to be aggregated into centralized repositories for model training. This approach raises privacy concerns, particularly in sensitive sectors such as healthcare. Federated learning offers a compelling alternative by allowing models to be trained where data resides, thereby enhancing privacy protections.
In this arrangement, AI models are deployed within existing hospital networks, reducing the need to move sensitive data. As Tripathi aptly noted, “We are not moving the data to the model. We are bringing the model to the data.” This infrastructure allows for the sharing of encrypted parameters and model updates across systems, ensuring that patient data remains secure within its initial environment.
Structuring Knowledge Before Model Training
Additionally, Tripathi highlighted the role of knowledge graphs in healthcare AI systems. Unlike large language models that are often the focus of AI discussions, healthcare applications require structured representations of medical knowledge. Knowledge graphs facilitate this by mapping complex relationships among genes, diseases, treatments, and outcomes—enabling AI to generate insights aligned with clinical reasoning.
This stage of system design is critical. According to Tripathi, “The real strength is not in the model. The real strength is in what happens before the model.” By integrating knowledge graphs with multimodal healthcare data—such as lab reports and patient histories—AI can deliver significantly more relevant insights.
From Data Pipelines to Agentic Systems
The need for efficiency extends beyond data collection to data preparation. Healthcare data often comes in inconsistent formats, leading to substantial manual labor in cleaning and standardizing this information for analysis. Agentic AI can streamline this process by automating many tasks, such as error detection and data standardization, enabling researchers to focus on analysis rather than data preparation.
Tripathi elaborated on the transformative potential of agentic AI, stating that it can drastically accelerate scientific workflows and improve data quality in the healthcare sector.
Implications for Drug Discovery and Clinical Research
The influence of AI technologies extends beyond administrative efficiency; it significantly impacts pharmaceutical research and drug development. Traditional drug development requires immense resources and time, with many clinical trials failing due to difficulties in patient recruitment or early detection of meaningful signals.
AI systems that utilize federated healthcare data can help identify patient patterns that enhance clinical trial design, thereby streamlining recruitment processes. These insights can ultimately lead to quicker drug discovery and broader patient access to experimental treatments.
Designing AI Systems for the Real World
In conclusion, Tripathi urged developers to focus not only on individual AI models or interfaces but also on creating integrated systems that combine various technological components. Effective healthcare AI will necessitate the orchestration of federated learning, knowledge graphs, and agentic systems to solve complex real-world issues.
By embracing these sophisticated architectures, healthcare institutions can better analyze fragmented medical data while preserving essential privacy. This shift could open new avenues for clinical research, drug discovery, and informed healthcare decisions.