Ali
Madani

He Came for Proteins. He Stayed for Everything.

In 2017, Ali Madani was staring at ultrasound images of beating hearts at the UCSF Cardiovascular Research Institute, teaching convolutional neural networks to classify echocardiograms faster than any cardiologist could. That paper - Fast and accurate view classification of echocardiograms using deep learning - now carries over 700 citations. It was not, however, what he had in mind for a career.

Madani earned his PhD in Applied Science and Technology from UC Berkeley between 2014 and 2019, won the UC Dissertation Award, and picked up an AHA National Innovative Research Grant along the way. He interned at IBM Research, building multimodal cognitive assistants for radiology and cardiology. He co-founded an EHR startup. But the thing that wouldn't let him go was a question nobody had seriously asked: What if you treated protein sequences like language?

At Salesforce AI Research, he got to ask it out loud. As the architect of ProGen - trained on 280 million amino acid sequences from over 19,000 protein families - Madani demonstrated that a large language model could generate functional proteins from scratch. Not approximate them. Not retrieve similar ones. Generate brand-new sequences with catalytic efficiency comparable to proteins that evolution spent billions of years refining. The 2023 Nature Biotechnology paper that came out of that work has 1,300+ citations. It's the kind of result that makes other scientists do a double-take and re-read the methods section.

He left Salesforce to build the company those results demanded. Profluent landed in Emeryville, California, with co-founder Alexander Meeske. The first investor conversation was not a formal pitch meeting. It was a Twitter DM - Madani reaching out to Nathan Benaich at Air Street Capital in 2021. That DM turned into Profluent's earliest backing and, eventually, the largest single position in Air Street's second fund.

400 Mutations Away From Anything in Nature

In April 2024, Profluent released OpenCRISPR-1 to the world - and it stopped biologists cold. Not because gene editing is new. Gene editing has Jennifer Doudna. It has CRISPR-Cas9. It has a Nobel Prize. What Profluent released was different: a Cas9-like protein designed entirely by AI, with no natural ancestor, 400 mutations removed from any sequence evolution ever produced. And it edited the human genome accurately.

The full technical paper landed in Nature in 2025: "Design of highly functional genome editors by modelling CRISPR-Cas sequences." Madani is the corresponding author. The result: several AI-generated gene editors showing comparable or improved activity and specificity versus SpCas9 - the gold standard that's been the workhorse of gene therapy for a decade.

Crucially, Profluent released OpenCRISPR-1 as open source, free to license for ethical research and commercial applications. That was a deliberate choice. Madani's vision isn't just to build proprietary proteins for paying customers. It's to prove that AI can democratize access to the tools of molecular biology.

The Nature paper received a NeurIPS 2025 spotlight for a companion result: scaling laws apply to protein design. The bigger the model, the better the proteins. This is the same dynamic that made GPT-4 better than GPT-3. Madani's bet is that it will do for biology what it did for language - compress decades of scientific progress into years of compute.

A Protein Design Engine Built for Scale

Profluent's stack runs deeper than any single breakthrough. The Profluent Protein Atlas is the world's largest private protein corpus - a proprietary dataset that no academic lab could assemble on its own. ProGen3, their frontier-scale protein language model, generates functional proteins across diverse families. E1 is a retrieval-augmented protein engineering model. Protein2PAM enables programmable gene editors with custom guide RNA targeting.

The commercial playbook is already in motion. Profluent has a multi-year collaboration with Corteva Agriscience targeting crop innovation, a strategic partnership with Revvity building base-editing systems, and a co-design partnership with Integrated DNA Technologies. These aren't licensing deals. They're bets that AI-designed proteins will work in the physical world, under real conditions, in real products.

Madani concentrates on one or two initiatives at a time - a leadership style that's unusual for a CEO managing 53 people and a $150M war chest. The company's stated aspiration hasn't changed since day one: dose the first human patient with an AI-designed therapeutic protein. When that happens - if it happens - it would be the moment where AI moves from writing code and generating images to literally healing people.

The First Patient Has Not Yet Been Dosed

Madani describes his goal with clinical precision: dose the first human patient with an AI-designed therapeutic protein. The milestones needed to reach that point - functional proteins, verified gene editors, commercial partnerships, scaling data - are either already checked off or actively underway. The $106M Series B announced in November 2025 is pointed directly at bridging the gap between lab breakthroughs and real-world clinical deployment.

Beyond therapeutics, the platform is expanding in three directions: agriculture (crops via Corteva), biomanufacturing (industrial enzymes), and diagnostics. The vision isn't one product. It's a platform where biology becomes programmable the way software is - where you specify what you need and an AI designs the molecular machinery to do it.

The company Jeff Bezos backed. The gene editor that doesn't exist in nature. The researcher who first asked: what if proteins have grammar? Profluent isn't a biotech startup with some AI on the side. It's what happens when someone with Madani's particular combination of machine learning depth and biological obsession gets the resources to find out if the bet is correct.

So far, the results say it is.