Translational AI for Health
Stratum is a NYC based applied AI lab focused on studying how AI capabilities become usable in real-world, health-related settings.
ABOUT US
Modern AI capabilities are advancing rapidly, but access to those capabilities remains uneven. In health-related settings, this gap is shaped by infrastructure, resources, and local context.
We are focused on closing that gap through translational AI research - building models, datasets, evaluations, and real-world systems to make AI usable in health.
WORK
Calm Journal: Local Language Models for Reflective Health Support
We explore whether small, locally running language models can support structured journaling and reflection without relying on cloud infrastructure. This project uses Calm Journal as a translational probe to study feasibility, responsiveness, and qualitative model behavior in low-stakes, health-related settings.
Automated Specialty Tagging for Medical QA Evaluation
We study the use of large language models as judges to automatically tag medical question-answering datasets by specialty, and evaluate how model performance varies across domains, model sizes, and model families. This work enables fine-grained analysis of capability and degradation in health-related AI systems.
AI Readiness and Health Infrastructure Constraints in New York City
We map the non-model constraints that shape where AI systems can realistically be used in health-related contexts, including electricity reliability, connectivity, hardware access, and institutional capacity. Using New York City as a case study, this work frames health disparities through the lens of AI deployability.
Resource-Constrained Decision-Making for Health Kit Assembly
We evaluate how AI systems reason and plan under explicit operational constraints by testing their ability to assemble health-related kits under fixed budgets, limited inventories, and verifiable requirements. This work studies failure modes, tradeoffs, and auditability in constrained decision-making tasks.