Neeraj Kumar, Pacific Northwest National Laboratory
This opening presentation sets the stage for TPC25’s comprehensive AI for Science program, providing attendees with a strategic overview of how artificial intelligence is revolutionizing scientific discovery. We will present the current landscape of AI adoption in scientific research, highlighting breakthrough applications from protein folding to climate modeling, and demonstrate why community-driven efforts like TPC are essential for advancing such transformation. The session will introduce participants to the carefully designed Hackathon and Tutorial programs, explaining how each track addresses critical aspects of AI implementation in scientific workflows. Attendees will learn about the progression from foundational concepts to hands-on implementation, including opportunities to work with large language models, build agentic systems, develop evaluation frameworks, and enhance research productivity with AI tools.
Vivek Natarajan, Google DeepMind
This talk highlights general purpose AI systems designed at Google to democratize medical expertise and accelerate scientific discovery. We will first take a look at the AI co-scientist, built to accelerate scientific breakthroughs by assisting scientists in generating novel hypotheses and aiding experimental design. This system has yielded validated results in areas like genetic discovery, drug repurposing, target discovery, and understanding antimicrobial resistance. Secondly, we will examine the AI co-physician, AMIE, developed to make medical expertise universally accessible through capabilities such as advanced diagnostic dialogue. In simulations, AMIE outperformed primary care physicians on multiple clinical evaluation axes and showed promise as an assistive tool, with ongoing real-world validations. Together, these AI initiatives demonstrate the potential to transform scientific research and care delivery.
This is a hands-on tutorial and hackathon for academic scientists with limited experience in agentic AI systems. Over 1.5 days, participants will learn to build and extend AI agents tailored to scientific challenges, particularly in biology and chemistry.
With guidance from mentors and access to NVIDIA and Cerebras compute resources, teams will collaborate on projects such as molecular tool development, protein engineering, and reasoning agents. This open-format event emphasizes collaborative learning and practical implementation, building on foundational AI concepts from a shared tutorial session. It is ideal for researchers eager to explore agentic systems and apply them to their own scientific work.
Participants will gain a working knowledge of agentic system architecture, learn how to apply agentic methods to domain-specific scientific problems, and develop prototype tools or agents. They’ll collaborate with peers and mentors, access advanced compute resources, and leave with hands-on experience that empowers further exploration of AI in science.
Shared foundations in AI for science
Intro to agentic systems and use cases
Team formation and project kickoff
Hands-on hacking with expert mentorship
Midpoint sync, debugging, optional breakouts
Final demos, project showcases, wrap-up discussion
Hackathon space is also being made available for working groups to use for internal planning and hands-on coding/testing. If you wish to reserve space for your team, please submit your request via email here. Put “HACKATHON” in the subject line and note the name of the working group, how many are expected to participate, and provide an abstract on the goals of your hackathon.
are open to TPC members and invited guests.
TPC25 is produced by TCI Media, home of the Wire Publications:
© 2025 TCI Media.
TPC25 is produced by TCI Media, home of the Wire Publications:
© 2025 TCI Media.