Transfyr Launches With $25M Seed to Record the Science That Never Gets Written Down

AI reads every paper ever published. What it cannot read is how an experiment actually worked — how many times it failed first, what the hands did to get past the finicky step, which instincts nobody ever wrote down. The scientific record is a lossy copy of reality. That is why moving a result from one lab to another takes months of training and troubleshooting, and often fails anyway.

Transfyr launched with $25 million in seed funding to fill that gap, announced on August 26.

The person who built ARPA-H, and the person who took Ginkgo public

The founders are an unusual pair.

Co-founder and CEO Anna Marie Wagner sat on the executive team at Ginkgo Bioworks, where she built the AI organization and ran every major strategic transaction — including the 2021 public offering that raised $1.6 billion, still the largest go-public raise in biotech history.

Co-founder and Chief Innovation Officer Dr. Renee Wegrzyn was the founding director of ARPA-H, the US health moonshot agency. Appointed by President Biden in October 2022, she designed the agency from scratch and was dismissed by the Trump administration in February 2025. She is now building in the private sector what she was hired to build in government.

It doesn’t automate the lab. It watches it.

Transfyr is not lab automation. It captures what a human actually does at the bench and converts it into machine-readable data.

Integrated sensor systems and multimodal models passively record operator actions and intent, environmental context, equipment telemetry and supply chain dynamics. That metadata surfaces sources of process variability, enables root cause analysis, optimizes protocols, generates training and tech-transfer SOPs, and produces robot-level instructions — with reinforcement learning loops for new protocols and environments.

Headquartered at The Engine in Cambridge, Massachusetts, the company runs its own wet lab, generating foundational training data, testing its sensor stack in real experimental workflows, and running evaluations for top frontier labs.

“Science is missing a critical layer of infrastructure that’s necessary for efficient reproducibility, translation, scaling, and automation,” said Wagner. “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.”

The 64% number

One figure in the announcement sizes the problem. Per an Accenture report, 64% of drug-launch delays in 2024 stemmed from chemistry, manufacturing and control (CMC) issues — with tech transfer a major component.

Tens of billions of dollars are written off annually on irreproducible research. A delayed medicine can cost a pharmaceutical company over a billion dollars. Patients go without. And successful handoffs still depend on apprenticeship.

“Breakthroughs mean nothing if they stay trapped in a single lab or depend on unwritten tacit knowledge to succeed,” said Wegrzyn. “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.”

A Nobel laureate and a transformer author on the cap table

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

The advisor list stands out: David Baker, the Nobel-winning protein folding researcher at the University of Washington; Jakob Uszkoreit, CEO of Inceptive Medicines and co-author of Attention Is All You Need; Chris Ré of Stanford; former Merck CEO Ken Frazier; Stanford biophysicist Stephen Quake; and Kevin Weil, former Chief Product Officer and Head of Science at OpenAI.

“Science is the world’s most important engine of progress, yet its ability to scale has lagged other critical industries,” said Hemant Taneja, CEO of General Catalyst. “This founding team brings a rare combination of scientific, technical, operational, and institutional experience to change that.”

The platform is already the core technology behind a nearly $1 million Massachusetts Life Sciences Center “Gamechanger” grant with the BioBuilder Educational Foundation, and features in a Genesis Mission program led by Boston University under the NSF’s $400 million Programmable Cloud Labs initiative.

The competition: the ‘AI scientist’ camp, and the data it doesn’t have

Most companies accelerating science with AI describe themselves as building an AI scientist. They operate at different layers, but the goal is the same: hand hypothesis generation, experiment design and interpretation to a model.

Periodic Labs is the clearest example. Founded by OpenAI’s former VP of research and Google DeepMind’s materials lead, it raised $300 million at seed last October and is building its own Menlo Park facility where robots run experiments at scale and AI analyzes the process and results — with the eventual goal of AI designing and running experiments itself. Discovery Loop, founded by Jeff Dean after 27 years at Google, aims to automate the research process by running thousands of experiments in parallel. Lila Sciences has raised $550 million for what it calls scientific superintelligence.

That is where Transfyr’s position becomes visible. The shared problem across this camp is training data. However many robot experiments you run, the protocol those robots follow was written by a human — and the protocol omits the handling and judgment that made it work. The trial and error Periodic Labs calls necessary before any great discovery is recorded nowhere.

This is why Transfyr runs evaluations for frontier labs and supports robotics research rather than competing with them. It is aiming to be a supplier, not a rival — which is what an observability layer for science actually means.

Renting out the lab is a separate camp. Emerald Cloud Lab, founded in 2010, has raised $92.1 million to let scientists design and run experiments remotely, and Chemify builds automated chemistry labs. Scispot raised an $8 million Series A for an AI-native lab platform, and lab data integrator Ganymede was acquired this January. All of them execute experiments; none captures the context of work done by hand.

Which means the company lives or dies on one question: can tacit knowledge actually be captured by sensors? Turning what nobody can put into words into training data is a compelling claim, but it is still a hypothesis rather than a product. The names on this seed round reflect the size of the bet, not proof that it works.