Mantis Biotech Builds Digital Twins of Humans to Solve Medicine's Data Problem

Eighty percent of clinical trials are delayed by data inaccuracies, and each loses an average of $15 million to problems as avoidable as manual entry errors and inconsistent measurement practices. Mantis Biotech, backed by Y Combinator and Pioneer Fund in a $6.3 million seed round, says it has built the infrastructure to fix that: a platform that converts fragmented biomedical and human-performance data into synthetic datasets powering validated digital twins of the human body.

The company describes its product as a "domain-aware data platform for life sciences" that encodes biological, clinical, and human-performance meaning directly into datasets. Rather than delivering raw tables that research teams must reinterpret for each new question, Mantis says it produces "canonical, reusable, continuously maintained datasets that reflect how life sciences work in practice." Those synthetic datasets represent anatomy, physiology, and behavior, the functional building blocks needed to simulate how a body responds to a drug, procedure, or environmental stressor.

Founder Georgia Witchel brings an unusual combination of credentials to the problem. A former elite ice climber who helped build the U.S. national team, she studied computer science and psychology at Harvey Mudd, computational mathematics at Johns Hopkins University, and biomedical engineering at the University of Washington. Before launching Mantis, she built Louiza Labs, a physics engine designed to power digital twins for autonomous robotic surgery and simulated FDA trials, giving her direct experience with the technical demands of regulatory-grade simulation.

Mantis claims to be "the first company that can access all data in a clinical trial through a full-stack search," a capability the company says is central to preventing the downstream errors that cause delays and cost overruns. That claim, alongside the 80 percent and $15 million figures, originates from the company's own platform materials and has not been independently verified.

The premise carries genuine weight. A continuously updated computational model of a patient or population could compress clinical trial timelines, improve dosing precision in rare-disease programs, and generate synthetic control arms where enrolling real patients is impractical or ethically fraught. But the value of any such model depends on how it is validated, what data it was trained on, and whether the gaps or biases in that underlying data are legible to the clinician or researcher relying on the output.

Mantis has not disclosed what specific data sources feed its synthetic datasets, how validation is performed, what privacy and compliance frameworks govern data handling, or whether any regulatory body has reviewed outputs from its platform. The company is actively recruiting senior data leaders in life sciences, a signal that it remains in early infrastructure-building stages rather than at commercial scale.

Whether digital twins become a backbone of clinical research or remain a compelling proof-of-concept will likely turn on the questions Mantis has not yet answered publicly: who owns the model when it is built from patient-adjacent data, who bears liability when a twin's prediction is wrong, and whether the auditability the FDA and EMA demand can actually be embedded into synthetic human data at scale. The $6.3 million gives the company room to start answering them.