Finding Right Patient At The Right Time: How Tanmay Sharma Is Advancing AI-Driven Patient Identification In Biopharma

In biopharma, success is often measured in clinical breakthroughs. But between a breakthrough therapy and a patient receiving it lies a quieter, more complex challenge, identifying the right patient at the right moment. Tanmay Sharma has built his career solving exactly this problem. This is precisely why we sought his expertise to better understand how artificial intelligence can transform patient identification and timing in real-world healthcare settings.

Currently serving as Senior Director, Market Intelligence & Decision Sciences at Corcept Therapeutics, and previously leading advanced analytics initiatives at Gilead Sciences, Tanmay has focused on one of the most difficult questions in healthcare: not just who the patient is, but when they become visible in data in a way that enables meaningful action. This distinction, between who and when, has defined his work across both HIV prevention and oncology.

A Problem Hidden in Plain Sight

In pharmaceutical analytics, patient identification is often treated as a segmentation exercise. Organizations look for diagnosed patients, map them to physicians, and build targeting strategies around those relationships. In reality, by the time a patient is clearly identifiable, critical decision windows may already have passed.

Tanmay’s work challenges this reactive model. Instead of focusing on confirmed states, his approach anticipates transitions, moments in a patient’s journey where clinical decisions are about to change. These moments are rarely labeled cleanly in data and must be inferred from patterns such as shifts in treatment, changes in testing behavior, gaps in therapy, and subtle signals across fragmented datasets. Turning these patterns into actionable insights is where artificial intelligence becomes essential.

Making the Invisible Visible in HIV Prevention

At Gilead Sciences, Tanmay worked on HIV prevention, where the central challenge is that the “patient” does not formally exist. Unlike oncology or chronic diseases, there is no diagnosis defining the target population. The goal is to identify individuals who may benefit from preventive therapy before they appear in traditional disease frameworks.

This makes the problem inherently ambiguous. The data does not explicitly indicate need. Instead, it offers fragments, such as testing frequency, healthcare interactions, and behavioral proxies, that may or may not signal risk.

To address this, Tanmay applied machine learning approaches suited for incomplete and uncertain labeling. Using positive-unlabeled learning, his models identified patterns among known relevant patients without assuming all others were irrelevant. Over time, this approach surfaced a broader, previously hidden population likely to benefit from prevention.

The result was not just a model, but a shift in how prevention analytics is approached, embracing ambiguity as a starting point rather than a limitation.

Detecting Rare Transitions in Oncology

If HIV prevention is defined by ambiguity, oncology presents a different challenge: rarity. At Corcept Therapeutics, Tanmay focused on ovarian cancer, specifically identifying patients transitioning to platinum-resistant disease. Within a large population, only a small fraction progress at any given time, often less than 2% per month, making early identification extremely difficult. This is not a static needle-in-a-haystack problem. The “needle” is a moving event, a transition that must be understood in the context of a patient’s evolving journey.

To solve this, Tanmay led the development of a patient-journey-based predictive framework. Instead of relying on isolated signals, the model reconstructs longitudinal patient histories, including diagnosis, lines of therapy, treatment gaps, biomarker activity, imaging patterns, and healthcare interactions. It then learns which sequences of events tend to precede progression.

Building for the Real World

A defining aspect of Tanmay’s work is its grounding in real-world constraints. Healthcare data is incomplete, delayed, and highly imbalanced. In oncology models, most patients do not transition in a given window, making it easy for standard algorithms to default to predicting no change. Tanmay incorporated ensemble methods and sampling strategies to ensure rare events were properly represented. Tree-based models were optimized using precision-recall metrics, aligning evaluation with real-world utility rather than theoretical accuracy.

Equally important, the models were designed for action. Predictions were not intended to remain in dashboards but to guide decisions. This required balancing sensitivity and precision, aligning outputs with field capacity, and translating patient-level insights into physician-level signals with context.

A Broader Shift in Biopharma

Across both HIV and oncology, Tanmay Sharma’s work reflects a broader transformation in biopharma. Patient identification is no longer just about classification. It is about prediction, timing, and context. It requires integrating diverse data sources, understanding longitudinal journeys, and designing systems that operate within real-world constraints.

Most importantly, it requires a shift in mindset, from analyzing what has already happened to anticipating what is about to happen. That shift is where the industry is headed. Through his work at Gilead Sciences and Corcept Therapeutics, Tanmay has been helping define what that future looks like.