In 2015, a PhD student named Tom Leung lost rare patient samples due to a faulty antibody.
It wasn't carelessness. The evidence to choose the right one existed somewhere in the literature. Tom just couldn't find it in time. No one could. The information was out there. It just didn't converge.
That's how BenchSci started—four co-founders, one failed experiment, one conviction that has never changed: the biggest bottleneck in drug discovery isn't scientific talent. It's the scientific infrastructure.
Ten years later, we're launching the product we've always been building toward.
Today, we're introducing EMET (Evidence Mapping and Exploratory Tool)—the Agentic Research Environment for preclinical R&D.
Modern biology is a fragmented nightmare. And it's getting worse.
The imperative to decode disease biology and bring medicines to patients has never been greater. New modalities are making more targets druggable than ever before. Genomic data is set to explode from 40 exabytes today to 500 exabytes by 2035. $400 billion in drugs are coming off patent in the next eight years, compressing timelines and raising the stakes on every program. The opportunity to understand biology at scale—and translate that understanding into drugs that work—is generational.
But the ability to actually do it? That's a different story.
Ask a scientist what their day looks like, and they'll describe a juggling act that gets harder every year. Thousands of data sources—genomic databases, clinical repositories, reagent catalogs, paywalled literature, internal ELNs, proprietary assay data—none of which talk to each other. Dozens of AI models, each purpose-built for a narrow task, none of which integrate. Thousands of software packages that require specialist knowledge to run. And an ever-expanding expectation of what a single scientist is supposed to know, produce, and decide.
What makes this particularly brutal is that it feeds on itself. Because we operate in a uniquely altruistic space—most tools and datasets in the life sciences are publicly funded and released for free—the volume of available resources keeps growing, with no one coordinating how they fit together. AI is now accelerating that explosion. More models, more data, more software, faster than any individual or team can absorb. The very progress that should be making discovery easier is making the infrastructure problem worse.
And the organizations where drug discovery happens aren't getting simpler either. They're getting more complex—more distributed, more cross-functional, more dependent on external collaborators, with higher regulatory bars and tighter timelines. The machine keeps demanding more of scientists while giving them fewer coherent tools to work with. As one R&D leader at a top pharma put it recently: "We're being tooled to death."
The result: scientists spend 40 to 60% of their time finding, cleaning, and organizing data. Not doing science. Not discovering. Managing infrastructure.
Every other technical profession solved this long ago. Software engineers got the IDE. Then they got Cursor—an AI-native workbench where an engineer and a team of expert AI collaborators work in a single environment. The cost of writing code collapsed, and with it, what's possible.
Biology never got its Cursor.
Until now.
Introducing EMET
Just as Cursor unified the software engineering stack, EMET unifies the scientific stack—data, AI models, software packages, code execution, and scientific reasoning in a single Agentic Research Environment purpose-built for preclinical R&D.
You ask EMET a complex research question in plain language. It decomposes it, identifies the right domain-specific skills, and chains them intelligently—across omics databases, closed-access literature, clinical repositories, and your own internal data. It invokes large frontier models and specialized biomedical models alike, selecting the right one for each task. It writes and executes code, runs computational biology workflows, synthesizes evidence across sources, and returns a cited, traceable, actionable answer. Think of it as an expert scientist in your specific domain, who also happens to be able to code, access every relevant database, and never sleep. It doesn't search. It reasons.
What lives inside EMET is a decade of work that cannot be assembled overnight. The Biological Evidence Knowledge Graph—858 million nodes, 2.2 billion relationship edges, the world's largest structured map of disease biology. Legal access to more than 16 million closed-access scientific papers through eight years of publisher partnerships—literature that no frontier model was trained on and no general AI can access at inference. The world's largest reagent and model systems database. Ontologies, proteomics, genomics, clinical data—all standardized and connected. And on top of that, a library of 200+ proprietary scientific skills and workflows encoding how expert scientists actually reason through the hardest problems in R&D.
It also connects to the world beyond the screen. EMET has a growing ecosystem of connectors that bring in external tools, data sources, and partner platforms—and direct lab-in-the-loop integration so that experimental design doesn't stop at insight, it flows through to execution.
The scientist stays in control—as the conductor of a team of agentic PhD scientists, not a passive recipient of AI output.
And it covers the entire preclinical journey: disease biology, target identification and validation, hit discovery, lead optimization, safety and toxicology, translational biomarkers, and IND-enabling studies. Every critical question, one environment.
What makes EMET different from every other AI tool in this space
This is where I want to be direct, because the "AI for drug discovery" space is crowded with claims.
First, something is important to say honestly: we're not trying to out-model OpenAI or DeepMind. The frontier models are extraordinary, and they're getting better every month. We use them. We build on top of them. We're not competing with them—we're the platform that takes the world's best models and puts a decade of scientific ground truth underneath them. The models are commoditizing. The biological context, the curated data, and the scientific reasoning that tells those models what to do in a drug discovery context—that's what isn't.
First: the data no one else has.
Eight years of publisher partnerships built legal access to more than 16 million closed-access papers—Elsevier, Springer Nature, Wiley, Oxford, and dozens more. Plus 780K patents, 310K preprints, and 1,000+ databases. No frontier model was trained on this. No general-purpose AI can access it. We can.
Behind the literature sits the Biological Evidence Knowledge Graph—our proprietary map of disease biology with 858 million nodes, 2.2 billion relationship edges, and 100 million ontological nodes, curated by 60+ BenchSci Ph.D. scientists. This is the ground truth of disease biology. It lives in EMET and nowhere else.
Second: scientific reasoning built by scientists, not software engineers who learned biology.
EMET encodes over 200 proprietary scientific skills covering every domain of preclinical R&D—exactly how a safety scientist reasons through a toxicity assessment, how a computational biologist approaches structural analysis, how a translational scientist builds a biomarker case. The reasoning is model-agnostic: EMET selects the best model for each specific task, whether that's a frontier LLM, a specialized biomedical model, or a purpose-built tool—because different models perform differently, and using the wrong one is as dangerous as using no AI at all. A neuro-symbolic evaluation loop delivers 95%+ accuracy on biological questions—2–4x better than frontier LLMs used alone, validated across 600+ tests and 8+ benchmarks.
Frontier models hallucinate biology. That's not a minor flaw—it burns up months of wet-lab time. EMET doesn't generate answers. It earns them.
Third: enterprise-grade from the ground up—and we mean that specifically.
In a recent conversation with a top pharma R&D team, someone put it plainly: their scientists want to upload files, but the organization needs to control what can be uploaded, by whom, and whether it persists after the session. That's one of dozens of governance questions that come up the moment a serious organization tries to deploy AI at scale. Who gets access to which connectors? How custom workflows are approved and versioned. Where data lives, how long it stays, and what compliance trail it leaves.
EMET was built for this reality, not retrofitted to it. Full integration of your proprietary and dark data alongside our assets. Feature-level governance controls so organizations can decide exactly what scientists can and can't do without going through engineering every time. Programmatic access, lab-in-the-loop integration, and a growing connector ecosystem—built for the complexity of how biopharma actually operates, not how a software company imagines it does.
Fourth: the scientific team that doesn't leave after onboarding.
Most vendors hand you a login and a knowledge base. We send in scientists. Every deployment includes dedicated scientific liaisons who understand your biology, your therapeutic areas, and your workflows. They run training, build bespoke workflows in hours, and manage change. This isn't support. It's a true scientific partnership and co-creation.
What scientists are already saying
We've been running early access with some of the world's leading biopharma companies—and through our free academic program, with scientists at universities and research institutions around the world. Here’s some of their feedback.
A scientist had spent months stuck on a target validation problem—every published protocol exhausted, cells failing, the project at risk. A single EMET session identified the hidden biological mechanism that no single protocol had captured, and proposed a path forward. The project was unblocked that same week. Months of stalled work to an experimentally validated result in days.
A safety scientist was assigned a deliverable that leadership had been told would require a dedicated team and many months to complete. One EMET prompt—the right data sources, the right reasoning—returned a structured, publication-ready output in under ten minutes. Her quote: "I haven't used a single person. In eight to ten minutes, I've got something. That's jaw-dropping."
80% of users report 25–60% efficiency gains. We've documented 50x time savings on specific workflows. Average direct cost savings per biopharma deployment: $6 million. These numbers come from scientists, not surveys.
What's surprised me most isn't the efficiency numbers. It's how EMET spreads inside organizations. Scientists pull other scientists in before we've formally rolled it out. In every account, it's gone viral. In a recent meeting with a top pharma organization, someone described it as expanding faster than any tool their company had ever adopted.
That happens because EMET was built by scientists for scientists—then engineered to work at enterprise scale. The order matters.
Why this matters beyond productivity
I want to be direct about what we're actually trying to do.
70% of drugs fail because we don't understand the biology well enough. Not because scientists weren't smart enough or didn't work hard enough. Because they were spending half their time managing infrastructure instead of doing science. Because the evidence they needed was buried in a paywalled journal, they couldn't access it. Because their AI tools made up mechanisms that sounded plausible but weren't real.
In a world where millions of papers are published every year, and thousands of new databases are released, the breakthrough can be a needle in a haystack: one critical node, one overlooked result, one fragment of a million-piece puzzle that changes everything. But that insight may be buried in a supplement, hidden in a subsection, or scattered across studies no scientist could manually connect. EMET makes the impossible findable—using powerful AI to surface the one piece that matters, accelerate the workflow, and help scientists move from searching to discovering.
EMET doesn't just make scientists faster. At its best, it makes the science more rigorous—more evidence-driven, more likely to catch the biology gap that would have killed a program two years and $200 million later.
If we do this right, fewer drugs fail. Fewer years are wasted. More patients get medicines that work.
That's what we started building in 2015 when Tom lost those patient samples. That's what we're launching today.
Drug programs fail when biology is misunderstood. EMET exists to close that gap—before it closes your pipeline.
EMET is now available. If you lead R&D at a pharmaceutical or biotech company and want to see it, request a demo.