Announcing our $9.22M seed round to build universal multimodal models for healthcare and an open research community
Today, we are excited to announce that Sophont has raised $9.22 million in combined pre-seed and seed funding rounds. Kindred Ventures led the seed round, which attracted a roster of distinguished investors including Google Chief Scientist Jeff Dean, Weights & Biases co-founder and CEO Lukas Biewald, Google DeepMind’s Logan Kilpatrick, Factorial Capital Partner and Hugging Face CEO Clément Delangue, Mei Ventures, Delphi Ventures, Upfront Ventures, Aiconic Ventures, Brightwing Capital, and Headwater Ventures.
At Sophont, we build multimodal medical foundation models—large-scale AI models trained on massive amounts of unlabeled clinical data—that can reason seamlessly across pathology slides, brain scans, clinical notes, and lab results. Our go-to-market is medical AI infrastructure: later this year we will release state-of-the-art model backbones that med-tech and pharma R&D teams can easily fine-tune, powering use cases such as triaging symptoms, discovering biomarkers, selecting patients for clinical trials, and more.
Our founding team brings deep expertise in the medical and AI domains. Tanishq Abraham (CEO) has more than six years of experience in medical AI, previously served as a Research Director at Stability AI, earned a PhD in biomedical engineering from UC Davis at just 19 years old, and founded MedARC, the world’s largest online medical AI research community. Paul Scotti (CTO) was formerly Head of NeuroAI at Stability AI and a postdoc at Princeton University, and brings over 10 years of experience in computational neuroscience. Together we’ve cultivated a close network of collaborators across dozens of academic institutions and medical centers where we’ve published open-source medical AI research in venues including Nature Biomedical Engineering, NeurIPS, ICML, and more. Unlike other companies in the medical space, our strategy is to embrace multimodality and open science.
A universal, multimodal foundation model for healthcare
Current medical AI is like the old parable of the blind men and an elephant. One group builds a great model for reading X-rays, another for analyzing DNA, and a third for parsing doctor's notes. Each feels a different part of the beast, but no one sees the whole elephant. A patient is not a collection of isolated data points; they are a whole person. Our multimodal approach means we're building an AI that can finally see the elephant.
We will first release unimodal foundation models for pathology (gigapixel slides), fMRI (functional brain scans), and clinical text, each tuned for its data-specific quirks and ready for immediate fine-tuning by partners. Next, we will combine our unimodal models together using late fusion architecture, aligning latent representations while preserving their domain-specific strengths, to enable cross-modal reasoning that surfaces novel biomarkers, drives personalized care, and scales with the flood data that defines healthcare. This roadmap allows us to plug seamlessly into diverse pharma and health-tech pipelines.
“In many areas of human knowledge, we’ve seen large generalized language models enable fascinating emergent capabilities, perceptive reasoning, and creative problem solving. However, in certain areas like medical science, there is a critical need for accuracy, reliability, and trust, where broader generalization may fall short. To this end, Tanishq and Paul possess a unique specialized model vision to bring domain-specific training data, architecture optimizations, performance and safety, and continuous improvement and trust through academic community and industry partnerships. At Kindred Ventures, we are honored to back the mission-driven Sophont founders as they work to make technological, pharmacological, and clinical impact at population-scale.”
– Steve Jang, Managing Partner, Kindred Ventures
Open science as a competitive edge
Sophont is the first medical AI startup built in true partnership with academics, clinicians, and the broader machine learning community. We are co-designing datasets, co-authoring studies, and releasing our models publicly. Such collaborations help us improve our models, deploy practically meaningful tools, and provide the transparency regulators and patients need to trust medical AI. We believe that the future of medical AI will be built on top of foundation models, and we are building and fostering the ecosystem for it. In a landscape of closed-source solutions and transactional relationships, our open-science ethos delivers a clear advantage: it nurtures long-term organic collaborations with clinicians and academic centers that naturally evolve into commercial partnerships.
Re-launch of our MedARC research community
In accordance with our open science aims, we are re-launching MedARC as Sophont’s open science research collective! MedARC was previously our public Discord server from when we were working at Stability AI—it is now owned by Sophont and will be the forum where we work alongside clinicians, academics, students, and hobbyists to build open models and publish scientific papers. While Sophont oversees funding, infrastructure, and commercial partnerships, MedARC operates as its independent R&D forum where anyone can follow our work in real time, contribute code or ideas, and earn co-authorship on peer-reviewed papers and top-tier conference submissions.
Volunteers can receive free access to Sophont-provided cloud compute, mentoring from expert researchers, and the chance to help train the next generation of multimodal foundation models for healthcare, all while retaining academic credit and the freedom to publish. We especially encourage academic PIs to contact us—your PhD student can pitch and steer a project that benefits from task delegation to the MedARC community and use of Sophont compute. You’ll retain first- and last-authorship and fast-track your research to high-impact publication.
This launch comes alongside revamped onboarding, brand new collaborative projects in pathology, neuroimaging, and clinical text, and a pool of free GPU compute ready for volunteers. If you want to help build the next generation of open medical foundation models—and publish while doing it—join us at https://discord.gg/tVR4TWnRM9.
We are hiring
The capital from our funding rounds will be used to secure GPU compute, expand data partnerships, scale up training runs, and hire a small team of exceptional researchers. Sophont is now hiring ML research scientists/engineers who share our conviction that open, multimodal AI will redefine healthcare, job postings at https://sophont.med.
- Tanishq and Paul
