Apoorva Srinivasan

Apoorva Srinivasan works on biomedical machine learning and exploratory interfaces for large language models. She contributed to PED-X-Bench, a collaborative project recording adult-to-pediatric extrapolation decisions in FDA drug labels, and developed curie, an interface experiment for navigating connected topics and resources. On her personal biography, she identifies herself as a researcher at Tatonetti Lab in Cedars-Sinai’s Computational Biology Department, studying drug toxicity and drug-drug interactions with machine learning.[2][4][3]

In her May 27, 2025 announcement of PED-X-Bench, Srinivasan described a dataset covering 737 FDA drug labels from 2007–2024. According to her account, the team classified extrapolation outcomes as Full, Partial, None or Unlabeled, alongside pediatric safety and efficacy summaries and metadata such as age bands, study design, indication and dosage form. A two-stage LLM generated annotations, and nine domain experts validated a stratified sample.[4]

Srinivasan argued that these records could support models distinguishing where adult evidence justifies on-label pediatric use from where child-specific trials remain necessary, helping direct resources toward children who need new studies. She credited TLab co-authors Nicholas Tatonetti, Jacob Berkowitz, Nadine Friedrich, Jose Miguel Acitores, Michael Zietz, Kevin Tsang, Aditi Kuchi, Hongyu Liu and Ryan Czarny.[4]

Before the pediatric benchmark announcement, Srinivasan publicly explored AI interfaces for learning. At the San Francisco AI | UX meetup documented in Latent Space’s April 26, 2023 recap, she presented a Personalized Generative Learning Graph, described as helping users identify what they needed to know and suggesting resources.[1]

In her November 21, 2023 essay introducing curie, Srinivasan described an LLM-interface experiment developed in her free time. She argued that search and chatbots can move users directly toward answers while bypassing opportunities for discovery. Applying the Physician/Librarian distinction she credited to Elan Ulendrof, she conceived curie as a “Librarian, inciting curiosity and enabling exploration.”[3]

Curie reveals subtopics connected to a user’s primary interest. Srinivasan illustrated this with a medical student learning machine learning: the interface helps expose how the subject is structured before the learner dives into individual resources. Nodes link to books, GitHub repositories, videos and articles, and open dedicated spaces for deeper exploration. Breadcrumbs preserve the navigation path, while personal nodes let users add their own insights to the knowledge canvas. She presented it as an initial experiment in moving beyond search and chatbot interfaces.[3]

Srinivasan’s earlier work provides context for her biomedical research. In her personal biography, she reports developing machine-learning algorithms for early cancer detection using glycoproteomics data at Intervenn, a liquid-biopsy startup. She also researched digital phenotyping of depression through passive smartphone sensor data at Columbia University. She holds a master’s degree in Biostatistics from Columbia and a degree in Biotech Engineering from the Manipal Institute of Technology.[2]