Transfyr launches with $25M in funding to bring physical AI to scientific labs

Transfyr has launched with $25 million in seed funding to tackle a problem that AI models trained on mountains of scientific literature still cannot see: much of what makes an experiment work never makes it into the scientific record.

General Catalyst led the round, joined by Lux Capital, Breakout Ventures, Factory, Neo, SV Angel, MVP Ventures, Underscore VC, Lyda Hill, and several angel investors.

The Cambridge, Massachusetts startup was founded by Anna Marie Wagner, former head of AI and corporate development at Ginkgo Bioworks, and Renee Wegrzyn, PhD, the founding director of ARPA-H. Transfyr is building what it calls an observability layer for science, using sensors and multimodal AI models to capture what scientists actually do at the lab bench and translate those activities into machine-readable data.

That distinction matters. AI can ingest papers, experimental results, protocols, and databases, yet the formal scientific record leaves out countless details that determine whether another scientist can reproduce an experiment. Failed attempts, subtle equipment adjustments, environmental conditions, operator decisions, workarounds, and hard-earned laboratory knowledge can disappear once an experiment becomes a paper or written protocol.

“Science is missing a critical layer of infrastructure that’s necessary for efficient reproducibility, translation, scaling, and automation,” said Wagner, Transfyr’s co-founder and CEO. “The existing scientific record is a lossy representation of reality, and we must build the interfaces that make the nuances of science observable and interpretable for future generations of scientists and the autonomous systems that will support them.”

Why investors are betting on the missing data inside labs

Transfyr’s thesis comes at a moment when AI companies are pushing models beyond text and software into machines that interact with the physical environment. Science presents an unusually valuable target. Laboratory work generates enormous amounts of information, yet much of the context around it has historically been difficult to capture at scale.

Transfyr deploys integrated sensor systems and multimodal models inside laboratories to record operator actions and intent, environmental conditions, equipment telemetry, and supply-chain information. The resulting data can help identify sources of process variation, troubleshoot failures, improve protocols, produce training materials, support technology transfers, and eventually create instructions detailed enough for laboratory robots.

That last piece could prove significant. Closed-loop autonomous laboratories need more than scientific papers and instrument readings. Machines need data describing how experiments are physically executed. Transfyr is betting that capturing this missing layer can help bridge human laboratory work with AI and robotics.

The size of the seed round and the people surrounding Transfyr suggest investors see that opportunity as infrastructure rather than another laboratory software application. Advisors and angel investors include Nobel laureate David Baker, Stanford professor Chris Ré, “Attention Is All You Need” co-author Jakob Uszkoreit, former Merck CEO Ken Frazier, Stanford biophysicist Stephen Quake, and former OpenAI chief product officer and head of science Kevin Weil.

“The real bottleneck to revolutionary science isn’t a lack of big ideas, it’s the massive friction of translating those ideas into reliable, scalable reality with impact,“ said Wegrzyn.

A costly commercial problem drives that friction. Transfyr cited an Accenture report estimating that 64% of drug-launch delays in 2024 stemmed from chemistry, manufacturing, and control issues, an area where technology transfer plays a major role. For pharmaceutical companies, failed transfers and manufacturing delays can turn scientific progress into years of lost time and substantial costs.

Transfyr is already working with organizations spanning diagnostics, academic research, workforce development, robotics, and frontier AI. Its technology is part of a nearly $1 million Massachusetts Life Sciences Center grant with BioBuilder Educational Foundation and a Boston University-led Genesis Mission program connected to the National Science Foundation’s $400 million Programmable Cloud Labs initiative.

The harder question is whether Transfyr can capture messy, highly specialized laboratory activity across many scientific environments and convert it into data consistent enough for machines to learn from. Labs, equipment, protocols, and human behavior vary enormously.

If it can, Transfyr could occupy an interesting position in physical AI: not building the scientist or the robot, but building the data layer that helps machines learn how science actually gets done.