AbInitio Bio is filing for priority on the science’s timeline, not its law firm’s.

Daniel Mukasa is the founder of AbInitio Bio, a YC Spring 2026 company building foundation models for biomanufacturing. He is mid-batch prepping for a raise, and negotiating speaking contracts at pharmaceutical companies and CDMOs (Contract Development and Manufacturing Organizations). In biotech, that kind of visibility is how trust gets built and how deals start. It's also where the trap is. Talk about a new idea before you've protected it, and you can lose the chance to protect it at all. So a small company's ability to file quickly and quietly decides how freely it can speak, share, and partner. File fast and you can say yes to the talk. Sit in a backlog and you're turning opportunities down.

This is a new problem for Mukasa. He did his PhD at Caltech and was a postdoctoral fellow at MIT and the Broad Institute. At world class institutions like these, filing a patent meant walking down the hall to one of the best IP teams in the country and getting it turned around fast. He filed his first patent that way, having handed over the materials, in his words, "probably as late as possible because I was super unorganized." It didn't matter. The lawyers locked in the date and moved on.

Then he left to start AbInitio Bio.

"When you leave these large institutions and their in-house legal counsel, you don’t have the same support anymore. Filing a patent application can take weeks if not months now because you're just another company in a Big Law firm portfolio. If you're a fast growing startup like us, that timeline could kill you.”

The companies that don't face this are the well-funded ones spun out of large venture studios, who buy their way around it with in-house legal teams. They file fast, so they can publish, disclose, and build trust at the pace the industry rewards. Everyone smaller has to weigh each opportunity against a backlog of paperwork.

Not a wrapper.

Mukasa had tried legal tools built on top of Claude and ChatGPT. He knew what they could and couldn't do. He knew the tell: confident-looking drafts that read fine but fall apart under scrutiny from a subject matter expert. Patent applications are a difficult format for pure-LLM systems because they are long, structured, internally cross-referential, and graded on the precise scope of claims rather than the surface fluency of prose. The interesting bits, the parts that determine whether the patent is enforceable, are exactly the parts a pure-LLM system is most likely to hallucinate.

Fearn is not a wrapper. Its co-founders, Han Kim and Angela Gao, PhD, built the system from the ground up: dozens of small LLM and non-LLM models, some built completely from scratch, composed together. The point of composing many small models rather than prompting one big LLM is control. Smaller models, LLM or not, are more controllable and resistant to hallucination, and when you compose them symbolically, you can produce patents at a level of reliability that no pure-LLM generation can match.

The deeper trick is the data. Most modern AI architectures assume a massive but imperfect training corpus and engineer around the noise. Several of the models Fearn built from scratch assume the opposite: a small, hand-corrected, hand-labeled dataset that is highly reliable rather than highly abundant. Designing a model that exploits that inversion, that gets more out of one carefully labeled example than a generic architecture would get out of a thousand noisy ones, is most of what Fearn's research has been about.

For Mukasa, the question was simple. He'd seen a top institutional team produce fileable work fast. Could anyone outside that walled garden do the same?

The test that matters.

The weekend draft was proof. A real patent application, built from technical descriptions that led to a full application draft. He refined it over the weekend, and had the autonomy to file whenever was best for him and the company.

That autonomy is the part that's hard to get at his stage. A strong draft you control, ready when you need it, means you decide when to file, when to speak, and when to share, instead of waiting on a queue to decide for you. The wrapper tools Daniel had tried never got close to producing something he could actually move on. Fearn did.

And a free piece of advice.

Mukasa recommends Fearn to technical founders broadly, and especially to anyone whose company depends on getting strong applications on file early, when the cost of moving slowly is highest.

"For a company my size, the hard part is getting a strong first draft together quickly. Fearn got me there in a weekend. That kind of speed, early, is hard to overstate when you're moving at startup pace." 

— Daniel Mukasa, Founder & CEO, AbInitio Bio

Building in biotech and need to file on a real timeline? Try Fearn at fearn.ai.