New York-based Mantis Biotech is addressing a critical bottleneck in biomedical research: the scarcity of high-quality data for rare diseases and edge cases. While large language models have transformed healthcare through structured data, they struggle when reliable information on unusual conditions is unavailable. Mantis proposes a solution by creating digital twins of the human body, which are physics-based, predictive models capable of simulating anatomy, physiology, and behavior. The company's platform aggregates disparate data sources, including textbooks, medical imaging, motion capture cameras, biometric sensors, and training logs. It utilizes an LLM-based system to route and validate these streams before feeding them into a physics engine. This engine generates high-fidelity synthetic datasets, allowing the creation of virtual humans that accurately reflect physical constraints. This approach enables the generation of data that does not exist in the real world, such as simulating hand movements for individuals missing a finger, thereby solving the problem of missing labeled data in specific populations. Georgia Witchel, Mantis founder and CEO, explains that this technology allows for the prediction of human performance under various conditions. A prominent early application involves professional sports, where the company already collaborates with an NBA team. By creating digital representations of athletes, Mantis can track daily metrics such as jump height, sleep duration, and lifting frequency to identify injury risks. For instance, the system can predict the likelihood of an Achilles injury based on training load and diet, demonstrating the potential to prevent issues before they occur. The significance of this technology extends beyond athletics into broader healthcare and pharmaceutical research. Digital twins could streamline clinical documentation, accelerate drug discovery, and support clinical decision-making. They offer a unique way to study rare diseases where ethical and regulatory constraints often prevent the use of real patient data in public datasets. Witchel compares the utility of these models to a child experimenting with toys, suggesting that testing medical procedures on virtual humans rather than real patients is a safer and more ethical alternative. To support these efforts, Mantis recently secured $7.4 million in seed funding. The round was led by Decibel VC, with participation from Y Combinator, Liquid 2, and various angel investors. The company plans to allocate these funds toward hiring, marketing, and expanding its go-to-market strategy. Looking ahead, Mantis aims to release its platform to the general public with a focus on preventative healthcare. Additionally, the company is developing tools specifically for pharmaceutical laboratories to assist with FDA trials. By providing insights into patient responses to treatments through simulation, Mantis intends to revolutionize how researchers approach clinical testing, offering a scalable method to overcome data limitations in medical innovation.