Saran Saund is responsible for the system in this talk: 28 million biomedical abstracts on Qdrant, refreshed daily, used in production to find drug targets that no single paper names. Two drugs found this way went from concept to clinic in 24 months, against an industry norm of four to seven years. He does not write the queries. He is the one who decided top-K was the wrong primitive, and he has spent a career making that kind of call: Xerox PARC, two companies with exits to Marvell and Qualcomm, then PointR Data, acquired by Oncotelic, where he now heads AI. He keynoted Qdrant's Vector Space Day in June 2026. His obsession is the unglamorous half of retrieval: everything that happens before anything gets ranked.