Francesco Cavalli

Project

Agentic AI for Therapeutic Strategy Discovery in Rare Diseases

This project investigates how agentic AI can support the discovery of therapeutic strategies for rare diseases, where clinical evidence is often sparse, fragmented, and difficult to translate into actionable development decisions. As biomedical innovation increasingly depends on integrating heterogeneous information across clinical studies, pharmacology, and disease biology, there is growing interest in AI systems that can do more than retrieve papers or summarize findings. Francesco’s project explores how agentic systems can help structure this evidence, identify therapeutically plausible opportunities, and generate research hypotheses that are more closely aligned with real translational constraints. Rather than only asking whether a model can produce a convincing idea, the project studies whether those ideas are grounded in available evidence, whether they remain scientifically coherent, and whether they can support more credible decision-making in early therapeutic development. Scientifically, the project pushes beyond conventional biomedical NLP and retrieval tasks by focusing on the quality of reasoning that AI systems apply to complex therapeutic questions. Rare-disease settings are especially demanding because useful signals are often distributed across small studies, heterogeneous endpoints, and incomplete evidence bases, making it difficult to move from isolated findings to structured therapeutic hypotheses. By studying how agentic AI can organize, connect, and reason across these fragmented sources, the project contributes to the broader research agenda around evidence-grounded generation, scientific decision support, and human–AI collaboration in high-uncertainty environments. In that sense, the work is not only about applying AI to biomedicine, but about building a framework for how intelligent systems can assist expert-driven scientific reasoning in domains where traceability, coherence, and judgment matter. The commercial relevance is especially strong because therapeutic development is often slowed by weakly structured opportunity discovery, large evidence gaps, and long cycles of manual evaluation before a viable development path can even be defined. This is particularly true in rare diseases, where programs must often be built from limited data and where strategic errors early in development can make approval pathways much longer, riskier, and more expensive. By helping generate hypotheses that are explicitly grounded in clinical evidence and oriented toward minimizing evidence gaps from the outset, the project points toward a more realistic approach to therapeutic strategy design—one that could support shorter, cheaper, and more defensible paths toward approval. In practice, this creates a strong foundation for AI-enabled infrastructure that helps biotech, pharma, and translational research organizations identify better opportunities earlier, prioritize them more rigorously, and enter development with a clearer evidentiary and regulatory rationale.