Serova is patenting every vaccine it designs — not just the one it can afford.

For a clinical-stage biotech, a deep patent portfolio used to be a luxury. The economics have changed.

[Founder name] started Serova to do what the cancer-vaccine field has been circling for a decade: use AI to design personalized vaccines faster than the disease can adapt. [She] came to it from [a computational immunology background — a PhD at [institution] and several years at [a leading vaccine lab]], where the science was the hard part and the paperwork was somebody else's job. Serova's platform doesn't produce a single invention. It produces a stream of them — the models themselves, the neoantigen vaccine designs those models generate, the manufacturing methods that turn a design into a dose. Each one is the kind of thing a competitor would copy the moment they saw it.

The trouble was never the inventing. It was the paper.

A patent draft, quoted by a top firm, runs about $10,000 — and in intellectual property, that is not a gouge. It is the market rate. It is also, for a company at Serova's stage, a forced choice. The future is unknowable, and a startup with a fixed IP budget has to guess which of its inventions will end up mattering, protect that one, and leave the rest in the open. The honest answer is to patent all of them. At $10,000 a draft, no early-stage company can.

And in cancer vaccines, an unprotected invention is not a risk you carry quietly. With Moderna, BioNTech, and a dozen well-funded startups all running at overlapping neoantigen targets, exposure isn't theoretical. It's a clock.

The second problem

There was a second problem, one [Founder name] knew from watching the disclosure process up close. The standard biotech patent workflow starts with a scientist explaining an invention to an attorney. Weeks later, a draft comes back, rewritten into legalese dense enough that you cannot easily tell whether the attorney grasped the underlying science or merely produced something shaped like a patent. The drafts always look plausible. Whether the claims actually capture the invention — whether they would survive a challenge — is a separate question, and not one a scientist can answer by reading the document.

When [Founder name] first heard about Fearn, [she] was skeptical for the usual reason. [She] had tried the legal tools built on top of Claude and ChatGPT. They all had the same tell: confident drafts that read fine until a lawyer looked closely, then fell apart. Patent applications are a hard format for a pure-LLM system — long, structured, cross-referential, and graded on the precise scope of the claims rather than the fluency of the prose. The parts that determine whether a patent is enforceable are exactly the parts a pure-LLM system is most likely to get wrong.

Fearn is not a wrapper. Co-founders Han Kim and Angela Gao built the system from the ground up: dozens of small models, LLM and non-LLM, several built from scratch, composed together analytically. Smaller composed models are more controllable and far more resistant to hallucination than one large model carrying the whole task. The deeper bet is the data — several of Fearn's models assume a small, hand-labeled, hand-corrected dataset rather than a large noisy one, and are designed to extract more from one carefully labeled example than a generic architecture gets from a thousand sloppy ones.

That was the first technical description [Founder name] had heard from a legal-AI company that sounded like it might actually hold up.

The test that matters

[Founder name] used Fearn to draft [her] first application [over a single weekend, ahead of a filing deadline [her] outside firm had said it couldn't meet]. An inventor describes the invention in their own words, the way they would to a colleague; Fearn reflects back, in real time, what it has understood — closing the loop on the question that the traditional process leaves open for weeks. The draft came back, and [Founder name] interrogated it through Fearn's chat, asking why each claim was written the way it was. Every answer pointed to the specific passage it referred to.

Then came the only test that counts. The draft went to Serova's outside IP counsel — the same firm working on the firm's timeline — and they reviewed it and filed it. They did not rebuild it. They did the light editing attorneys do on their own work, not the teardown they do on a draft they don't trust.

Serova has since drafted several more applications with Fearn, each reviewed and filed by outside counsel. The portfolio is growing at a pace that, until recently, was not available to a biotech at Serova's size.

"In therapeutics, one patent can be the whole company. Fearn lets us protect every invention worth protecting — not just the one the budget allows — and file each on the science's timeline instead of the firm's."

— [Founder name], Founder & CEO, Serova

If you're building in cancer biotech and racing to file before the field catches up, try Fearn at fearn.ai.