Xlue, a Carnegie Mellon University spinout building a foundation model for human health, has raised a $35 million seed round led by Conviction, Forbes reported Nov. 18. Pranav Reddy, a partner at the firm that Sarah Guo founded, led the investment, according to Forbes' InnovationRx newsletter.
The financing comes as investors back what are being called clinical foundation models: large AI systems trained on medical data rather than general text. Menlo Ventures' 2026 state of AI in healthcare report places Xlue and a company called FOCI in that category.
What Xlue says it is building
On its website, the company describes an "intelligence layer" meant to connect healthcare applications with medical data. Intended uses include preventive care, clinical decision support, population health and pharmaceutical development. Its Nov. 18 announcement said prospective studies, along with drug-indication work, are underway with hospitals and pharmaceutical companies. It named none of them and offered no results.
The product's lineage is a little tangled. Earlier work centered on CATCH-FM, a cancer-risk model built on next-token prediction, the same training principle behind chatbots, according to Carnegie Mellon. The November launch leans instead on the phrase "human health world model," and an October evaluation refers to the research model as HealthSAW. These may be successive stages of one effort rather than the same product, though the company has not said so.
Carnegie Mellon said the team built its own model rather than adapt an off-the-shelf large language model, partly because the records were nonpublic and patient privacy was a priority.
How much the evidence shows

Protege, a healthcare-data company, published a double-blind evaluation of HealthSAW on Oct. 21. It measures retrospective disease-risk prediction, meaning how well the model flags conditions in records of what has already happened. According to the evaluation materials, it does not show that treatment simulations improve trial success, that clinical use improves patient outcomes, or that deployment lowers healthcare costs.
Protege's report also estimates that prediction costs 11 to 900 times less than with frontier language models. That figure rests on model size and reference pricing, not measured deployment spending, and it leaves out integration, operations and follow-up care. Xlue's benchmark comparators included LightGBM, XGBoost and frontier language models.
The company has been careful in how it describes the technology. "This is not diagnostic, it is risk stratification. This is about who the doctors should test early," a company representative told Forbes.
Reddy, for his part, characterized the approach as forecasting health trajectories rather than isolated symptoms, and credited the team's experience training and running large models in production, according to a post on teapawt.org.
The founders and the money
Carnegie Mellon identified Chenyan Xiong, Hao-Ren Yao and Ophir Frieder as the original founders. Xiong earned his Ph.D. there in 2018, according to the university's alumni record, and worked part time at Meta in 2024 on recommendation foundation models. Frieder is a Georgetown faculty member and also serves as chief scientific officer of Invaryant, according to his faculty page.
A Form D filed with the Securities and Exchange Commission identifies Xlue as a Delaware corporation formed in 2025 with a Pittsburgh address. It reported $450,000 sold against a $5 million offering to two investors. That is far smaller than the $35 million round, and the filing's relationship to the seed is not clear from the record.
Earlier, Technical.ly reported Feb. 5 that the company had raised $1.5 million through a SAFE, a simple agreement for future equity, with the valuation undisclosed. PitchBook lists Embedding VC, Transpose Platform Management and Carnegie Mellon as investors, but its profile does not establish that they joined the November seed. Dealroom shows an enterprise-value range of $140 million to $210 million, though the page does not say that is the round's valuation. Xlue's valuation has not been confirmed.
Headcount is similarly fuzzy. The Pittsburgh Post-Gazette reported six employees in February, while Xlue's LinkedIn page shows 2 to 10.
Rivals and early partners
Xlue is not alone in the field. Knit Health announced a separate $35 million Series A on Nov. 16, led by NEA, for its Med-BEAM healthcare world model. It is a different company at a different stage.
In September 2025, Xlue signed a strategic partnership with Kaohsiung Medical University Chung-Ho Memorial Hospital in Taiwan to bring its model into clinical workflows, starting with cancer prescreening. "By partnering with KMUH, we are bringing our early disease prediction solution directly into hospital practice," the company said in that release. Technical.ly also reported in February that Xlue was in talks with UPMC over access to patient data. Neither arrangement amounts to a confirmed commercial deployment.
What comes next

Prospective clinical proof, the kind that tests predictions on patients going forward rather than on past records, is still pending. Xlue says such studies are underway, but until it names its hospital and pharmaceutical partners or publishes results, the $35 million rests largely on promise.
