Fabio Urbina, Associate Director at Collaborations Pharmaceuticals, shares his career journey from cell biology to machine learning-driven drug discovery. He explains how his company uses ML models to virtually screen chemical compounds for rare and neglected diseases, dramatically narrowing down candidates before expensive lab testing. A key highlight is their work building predictive models from as few as 15 data points using prototypical networks — an embedding-based approach related to k-nearest neighbors — running on modest hardware. Fabio also discusses the challenge of small, narrow training datasets in drug discovery versus the massive datasets used in LLMs, and why classical models like SVMs and random forests often outperform newer architectures in data-scarce domains. He closes with career advice about embracing discomfort when switching fields.