Hiding in Plain Sight
Hey there, investors and innovators!
This week's slate has a theme that's easy to miss at first glance: none of these five companies are trying to generate new information — they're trying to finally read information that already exists. A biotech founder realized the immune system had already solved the vaccine design problem; a legal-tech team realized the paper trail for whistleblower fraud was already sitting in unsealed court dockets; a computer vision founder realized the eyes for the physical economy were already installed on a billion-plus cameras. Across law, biotech, insurance, physical-world AI, and AI training data itself, this batch is less about inventing new capability and more about finally extracting the signal that was always there... and building the traceability to prove it.
Eight Capital tracks every YC batch, cohort to cohort. Reach out if you'd like a deeper cut on any theme above.
Fund Manager Picks
TareBio
rewriting how vaccines are discovered.
Why it's interesting
For 200 years, vaccine discovery has been a prediction problem: guess the right antigens. TareBio makes it a readout problem — capture the immune cells already fighting a disease and reverse-engineer the vaccine from them. They've already cured lymphoma in mice with a platform that generalizes across cancer and infection, led by a decade-deep MD-PhD scientist-founder with a Gates-funded origin and a world-class advisory board.
Veltha
Regulation as the moat
Why it's interesting
The "regulatory decision layer" for insurance claims, starting with crop and workers' comp. It reads every document in a claim and checks it against the actual federal rulebooks, producing paragraph-level citations an adjuster will sign — turning a six-hour review into ten minutes. Two ex-Palantir founders, exactly the forward-deployed DNA for embedding inside a regulated enterprise.
OpenVector
a digital coworker for the physical economy
Why it's interesting
Everyone builds agents for knowledge workers, but ~80% of the economy is physical. OpenVector's agent understands any real-time camera feed — from the billion-plus already installed — and acts on it, no new hardware required. The technical founder spent six years at the Weizmann Institute under Shimon Ullman with a CVPR spotlight to his name.
Erinys
AI-native litigation against fraud.
Why it's interesting
Erinys compresses months of pre-filing investigation into days for False Claims Act cases — the whistleblower suits that recovered a record $6.8B last year. It drafts the complaint with every assertion traced to source, then corroborates outward across dockets and federal payment data. The moat is proprietary: they've indexed every unsealed federal qui tam docket from 2010 to 2026.
Datoric
The data layer for the next models.
Why it's interesting
Security-first, fully licensed training data for voice, robotics and world models — multilingual speech, egocentric video and action trajectories, all sourced with consent and clean licenses. The founders learned the game by beating it first, then built the defensible version. Roughly seven figures of revenue in the first 30 days.
