The complaint is familiar to anyone who's worked in pharmaceutical regulatory affairs: somewhere, buried in the FDA's sprawling digital archives, is the answer you need. Maybe it's in a guidance document from 2018. Or an approval letter. Or a review memo filed under an obscure therapeutic area. Finding it, however, means wading through thousands of PDFs, often with nothing more sophisticated than keyword search and hope.
Rhizome AI, a recent Y Combinator graduate, is wagering that generative AI can cut through that thicket. But here's the catch, and it's a big one in an industry allergic to guesswork: every answer has to come with citations. Not vague references—actual page numbers that link directly back to source paragraphs.
The company's pitch is deceptively simple: "Know what the FDA thinks." Early results suggest at least some corner of the industry is listening. According to metrics the startup displays publicly, Rhizome has signed up more than 20 customers since launching earlier this year, a modest but notable foothold given the conservative pace at which pharma and biotech typically adopt new tools.
The Technical Bet
Rhizome's platform doesn't just skim FDA documents. It ingests them—up to 1,000 per query, founder Chetan Mishra said in a recent interview. The underlying corpus spans roughly 2.5 terabytes and pulls from as many as 30 regulatory and clinical datasets across six global markets: FDA, European Medicines Agency, the UK's MHRA, Japan's PMDA, among others.
Under the hood, the system marries a large language model with what Mishra describes as a "proprietary retrieval and validation layer" tuned specifically for life sciences. The claim that stands out, perhaps more than the founders expected: "no hallucinations reported" since the product went live. That's a bold assertion in a field where a single unverified AI-generated sentence can torpedo a multimillion-dollar regulatory submission.
How much of that is careful design versus limited usage remains an open question. The company says it's generated over 2,800 answers since launch—a figure that suggests real activity, though hardly the scale where edge cases start surfacing in volume.
Pricing follows a tiered model. A free version covers 15 datasets with weekly updates. Paid access starts at $400 monthly for single-project use, climbing to $20,000 annually for business tiers that unlock broader datasets. Enterprise customers—who get South Korean, Chinese, and Japanese regulatory data, plus biweekly office hours—pay undisclosed rates.
Who's Using It?
Rhizome isn't naming customers publicly, but testimonials on its site reference users at clinical-stage biotechs, mid-tier commercial pharma, and what the company characterizes as a "Top-5 Pharma" client. LinkedIn posts from Mishra earlier this year noted 18 organizations using the tool; that count has since ticked up, suggesting steady if unspectacular traction during the Y Combinator batch.
The startup has also begun sponsoring regulatory affairs conferences and publishing analysis pieces dissecting specific FDA guidances—moves that feel less like marketing and more like building credibility in a community that doesn't hand out trust lightly.
The Founder's Pedigree
Mishra comes to this with a resume steeped in AI infrastructure. He was the first product engineer at EvolutionaryScale, where he helped scale a platform to tens of thousands of developer-scientists and billions of API calls. Before that, he was employee number 16 at Instabase, where he reportedly closed $7 million in enterprise deals. He's also a co-author on the ESM3 paper that landed on the cover of Science—credentials that carry weight in technical circles, if not necessarily with regulatory affairs veterans.
His stated vision: compress the cycles teams spend manually hunting through documents and synthesizing regulatory positions. The roadmap hints at workflow integrations beyond search, though details remain thin. Which might be intentional—overpromising in this market is probably worse than saying nothing at all.
A Fragmented Landscape
Rhizome enters a regulatory intelligence space already occupied by entrenched players: Clarivate's Cortellis, Redica Systems, POLITICO's AgencyIQ. These platforms bundle FDA tracking with broader clinical and commercial intelligence, and they're not cheap. Industry forums are littered with complaints about cost and fragmentation, which may crack the door open for leaner, AI-native alternatives.
Interestingly, the FDA itself rolled out "Elsa," an internal generative AI assistant, mid-last year. That the regulator sees value in AI-driven document analysis is notable—whether it validates the vendor market or simply raises the bar on what "good enough" looks like is harder to say.
The Trust Question
For now, Rhizome's value proposition boils down to speed, verifiability, and price. Faster answers. Clickable citations. A fraction of what legacy platforms charge. Whether that's sufficient to pry regulatory teams away from vendors they've relied on for years hinges less on technology than on something harder to engineer: trust.
In an industry where a missed citation or misinterpreted guidance can delay approvals—or worse, invite FDA scrutiny—being right matters more than being fast. Rhizome's early customers are, in effect, beta testing that promise in real submissions. The results of those experiments, one way or another, will matter far more than any demo.
