YC W25 Delineate builds AI agents for accelerated clinical trial design Custom LLM + computer vision reads data straight out of research figures Working with two top-10 pharmaceutical companies NIH STTR grant with MIT for quantitative systems pharmacology Months of evidence aggregation, done in weeks Founded 2024 in Cambridge, Massachusetts YC W25 Delineate builds AI agents for accelerated clinical trial design Custom LLM + computer vision reads data straight out of research figures Working with two top-10 pharmaceutical companies NIH STTR grant with MIT for quantitative systems pharmacology Months of evidence aggregation, done in weeks Founded 2024 in Cambridge, Massachusetts
Company Dossier · Cambridge, Massachusetts

Delineate

Quantitative evidence for consequential decisions.

An AI company teaching machines to read the entire scientific record - text, tables, and the numbers hiding inside charts - so pharma teams can design better trials, faster.

AI Drug Development YC W25 Seed Founded 2024
Delineate logo - a hand cradling a molecule
A hand cradling a molecule. The whole pitch, drawn small: put quantitative evidence in the palm of the people deciding which drugs advance.
15x
Faster processing
2
Top-10 pharma clients
90%
Of drugs still fail
$1-5M
Value of one saved day

Drug development is, at bottom, an enormous reading assignment. Delineate decided to automate the reading - and to notice that most of the useful data was locked inside pictures.

Here is a fact about the pharmaceutical industry that sounds like a rounding error and is actually the whole story: roughly 90% of drug candidates fail. Companies spend a decade and a fortune to learn, most of the time, that a molecule does not work. The obvious response is to make better decisions earlier, and the not-obvious problem is that the information needed to make those decisions is scattered across hundreds of papers, patents, and trial reports, much of it written by competitors, much of it decades old, and a surprising amount of it printed as dots on a chart that no keyword search will ever find.

Someone has to gather all of that. Traditionally that someone is a highly trained scientist, and the gathering takes months. This is the bottleneck Delineate walked up to and decided was, in fact, a business.

Delineate builds AI agents that extract and structure data from the scientific literature at scale. That much you could say about a dozen companies. The interesting part is what the company chose to obsess over: the figures. A lot of the most valuable numbers in a research paper never appear as text. They exist only as a curve, a scatter of points, an error bar on a plot. Delineate pairs custom large language models with computer vision that reads those images the way a trained analyst would, and then - this is the part that matters in pharma - runs the output through a quality-control process, because a confidently wrong number is worse than no number at all.

The value isn't the model. It's the quality control wrapped around it. In drug development, trust is the product.

The result, the company says, is roughly a 15x improvement in processing speed versus industry standard, and studies that shrink from months to weeks. That compression is not a vanity metric. In an industry where saving a single day of development can be worth $1 to $5 million, speed is a strategy. Delineate's pitch to a pharma customer is not "we have a clever model." It is "we will hand you, in weeks, an evidence base that would otherwise have cost you a team and a quarter - and every downstream decision gets sharper because of it."

The technical name for a lot of what this enables is model-based meta-analysis, or MBMA, a standard tool of modern drug development. MBMA leverages published summary data alongside a company's internal data to inform decisions like benefit-risk assessment and comparative efficacy - the kind of judgment calls that determine whether a program advances. The catch, always, is that MBMA starts with a human reading everything. Delineate automated the reading, not the judgment. The scientists still decide. They just decide with more evidence, sooner.

What Delineate is selling, in other words, is not a replacement for expertise. It is leverage on it. The company's own tagline - "quantitative evidence for consequential decisions" - is almost aggressively unglamorous, which is a point in its favor. It is not promising to cure anything. It is promising to make the people who might cure something less blind.

Three jobs Delineate is built to take off a scientist's desk.

Aggregate evidence

Pull structured data from thousands of papers and patents - including numbers embedded in figures and plots - and get a clean dataset in a fraction of the usual time.

Power MBMA

Assemble the evidence base behind model-based meta-analysis and model-informed drug development: benefit-risk, comparative efficacy, the totality of evidence.

Accelerate QSP models

Under an NIH STTR grant with MIT, speed up quantitative systems pharmacology model development and repurposing.

The bottleneck, drawn as a bar chart. Longer is slower.

Manual review
Months
With Delineate
Weeks
Speed multiple
~15x faster

Figures are company-reported approximations, illustrative of scale rather than exact benchmarks.

A pharma insider and a defense-AI engineer. The scars plus the chops.

EN

Emily Nieves

Co-Founder & CEO

MIT PhD who did computational research at Pfizer and AstraZeneca. She felt the evidence-aggregation pain on the bench, then left to build the tool she wished she'd had.

JI

Jawad Iqbal

Co-Founder & CTO

Robotics and AI background; previously led AI work at Lockheed Martin's Skunk Works. Brings frontier machine-learning engineering to a domain that punishes sloppiness.

"We are helping one customer get to the next phase of trials faster by creating the largest dataset on a drug class ever constructed."

Delineate, on its work with a top-10 pharma company

A short, fast first year.

Small details that stuck with us.

The logo is a hand cradling a molecule - evidence, literally handed to the decision-maker.

Its computer vision extracts numbers that only exist as pixels in a chart, never as text.

Co-founders came from Pfizer/AstraZeneca and Lockheed's Skunk Works - benches and black projects.

Talks, demos and interviews - search these out to see Delineate in motion.

Website, socials and the paper trail.

Delineate · Cambridge, Massachusetts · AI for evidence-driven drug development