We are extending the Jupyter ecosystem for collaborative, computationally intensive and AI-powered biomedical research
We are launching BioJupyter at the Berkeley Institute for Data Science (BIDS), as a component of the recently announced Bakar Computational Biomedicine Initiative (BCBI). With this effort, and in close partnership with BCBI scientists, we will extend the existing capabilities of the Jupyter ecosystem to support collaboration in the biomedical sciences — spanning the full range from basic biological research through clinical work. This effort will start in the BCBI community of research labs across Berkeley and UCSF, but will develop open infrastructure for scientific researchers everywhere.
Built on the open architecture and ethos of Project Jupyter, BioJupyter will be a modular, open platform for biomedical discovery — one that makes fluid, reproducible collaboration across teams and institutions a natural part of the research environment, with shared data, integrated AI, and full respect for data privacy and security policies.
Bringing seamless interoperability to Jupyter deployments
Today, biomedical scientists across UCSF and Berkeley report a number of important challenges with their current environments and workflows. They navigate complex hardware orchestration, the fast-changing capabilities (and pitfalls) of AI-assisted data analysis, and the adaption of bespoke, domain-specific tooling to new research directions. These bottlenecks, which impede within-team efficiency as well as collaboration, are multi-faceted and can't be solved magically with any one tool. Our vision is for BioJupyter to accelerate research and facilitate collaborative work across groups by deploying common, shared environments in partnership with research teams, while working with their scientists so they can share both their data and their code and workflows.
We do not aim to build (yet another) centralized repository for data. Instead, we will build new capabilities in JupyterHub, JupyterLab, JupyterAI and other subprojects in Jupyter, as well as plugins and integrations with existing biomedical systems to access them from these BioJupyter Hubs. JupyterHub is already widely adopted by many of our partners, but today's architecture treats each deployment independently, and cross-deployment collaboration requires manual and brittle processes. New work done in BioJupyter will enable researchers to seamlessly build collaborative projects across deployments, connecting teams within and across institutions with minimal effort, while retaining data privacy controls and oversight over computational resources in each setting.
Beyond Berkeley and UCSF
Our initial target is to facilitate collaboration across labs at UC Berkeley and UCSF in the Bakar ecosystem, and our team will work closely with researchers in those teams, and with the cohort of new faculty and postdoctoral scholars hired as part of BCBI. But everything we develop will be built as part of the open source ecosystem that Jupyter belongs to, so that researchers well beyond Berkeley and UCSF can collaborate with us, participate in the development, and share the tools we build. We expect engagement and collaboration with other organizations that have similar needs. We are actively working on some early collaborations of this kind, and will be announcing more soon.
All our work aims to strengthen Jupyter and related projects: we will upstream all relevant work into the core Jupyter subprojects above, so that scientists in other disciplines who share these needs around data sharing, collaboration, and the integration of AI into scientific workflows can also participate in the development and benefit from the results. And (in the spirit of communities like Pangeo), we will strive to upstream any advances we make to external projects, collaborate with their communities, and prefer to leverage and support those teams and technologies when possible. Building from the needs of BCBI researchers, we aim to have broad impact across the entire scientific open source ecosystem.
How BioJupyter relates to JupyterHealth
While separate from our existing JupyterHealth work, there are clear synergies between both: the architecture and community of Jupyter, BIDS' open science commitment, and the unique scientific environment of Berkeley and UCSF. JupyterHealth focuses on deep integration between patient-generated data and the real-world health care systems used by practicing clinicians, all on fully open source infrastructure; BioJupyter will support the cross-institutional, computationally intensive and AI-powered research aspects of biomedicine.
Fully open work, like the rest of Jupyter
Openness is central to our entire vision for this effort, anchored both in the values of Project Jupyter and in how our teams at Berkeley and UCSF work, and it shapes all four of the roles we are now recruiting for. We recognize that commercial and proprietary systems are important in today's AI-assisted research, and there is obviously an enormous flurry of industry activity in this space. But for us, science is a foundational layer of an open society: knowledge is essential to humanity being able to thrive, and we believe that Jupyter has a role to play here. What we now explore is how Jupyter can best enable scientists to use AI systems as an integral part of their research, while remaining grounded in open source tools and infrastructure, and with equal access to open and proprietary models as their needs dictate.
We do not imagine a future in which the only point of AI is to accelerate the production of publications, as we know the publication and peer review systems
are already straining under the load of AI-assisted submissions. What interests us is how scientists can bring AI tools into their workflows, adapt them to their own needs, and build research that brings meaningful value to society, from basic knowledge to health.
There is precedent for this: we saw exactly that kind of creativity with the development of the modern scientific computing ecosystem in Python and R, and a decade or so later in Julia. Those ecosystems were not simply cheaper versions of proprietary products — they were genuinely different. They allowed scientists to reimagine what their science could look like and to build things that vendors were not providing. We think that vision of the future matters enormously in the context of AI for science.
Finally, Jupyter is not in the business of developing AI models. But we want to facilitate first-class interactions with open models and open systems, which can offer a better fit for scientific workloads, and better privacy and environmental characteristics, than proprietary frontier models run only inside companies.
Members of the UC Berkeley community (and our JupyterHealth collaborators at UCSF, The Commons Project, and 2i2c) at JupyerCon 2025
We are hiring — come build this with us
The world of science today is not the world of science of 2010 or 2015, when the IPython (later Jupyter) Notebook was developed, or when JupyterHub was built. We want to imagine what those foundations can help us build to meet the needs of science for the next decade. Ultimately, we want to offer the scientific community — with an emphasis on biomedical research — the kind of tools that Jupyter provided ten or fifteen years ago with notebooks, JupyterHub, and our earlier infrastructure, but built for the modern age of cloud-native, AI-assisted science: deeply distributed, highly collaborative, and spanning very different types of infrastructure, from local clusters through HPC facilities all the way up to cloud deployments.
We are now recruiting four people at BIDS to build it: three open source research software engineers — working on computational infrastructure, on JupyterLab and user experience, and on agentic AI workflows in science — and one focused on scientific product management to bridge what biomedical researchers need and what we deliver for them.
Read the full job descriptions and how to apply.