The man who handed biology a "run" button.
AlphaFold can predict the shape of a protein. RFdiffusion can invent one from scratch. OpenFold, Boltz, the whole alphabet soup of molecular AI - any of them could shave months off a drug program. There is one catch, and Deniz Kavi says it out loud: "Most chemists and biologists are not programmers and can't write code to use these tools." The models are open. The doorway is not. Kavi's company, Tamarind Bio, is the doorway.
Today he is co-founder and CEO of a San Francisco startup that, in February 2026, raised a $13.6M Series A led by Dimension Capital. The platform he runs serves more than 10,000 scientists. Eight of the top twenty biopharma companies on earth - names like Bayer and Boehringer Ingelheim - log in to design antibodies, peptides, enzymes and radiopharmaceuticals. They do it through a web page. No cloud account. No SLURM cluster. No machine-learning PhD required. That is the entire point.
Our mission at Tamarind Bio has always been to remove the barriers between scientists and the most powerful computational tools available for drug discovery.- Deniz Kavi
It started with one annoyed lab.
Tamarind was not born from a pitch deck. It was born from a chore. As a computer-science undergraduate at Stanford doing computational-biology research - and later as a software engineer at Stanford University School of Medicine - Kavi watched the same scene repeat. Wet-lab biologists had a hypothesis. The computational scientists had the models. In between sat a swamp of manual setup: spinning up GPUs, wrangling dependencies, babysitting jobs that ran for hours. Brilliant people spent their days being IT support for AlphaFold.
He and his co-founder Sherry Liu - also Stanford computer science, also a stint at AWS - decided the swamp was the business. In 2023 they built a tool for a single Stanford lab. Researchers could click a model instead of configuring one. Word traveled the way useful things travel in science: fast and by mouth. The fix for one lab became a website for thousands.
With AI becoming a core part of the drug discovery R&D workflow, we started Tamarind two years ago to enable any scientist to design drug candidates, without thinking about cloud, data or AI inference infrastructure.- Deniz Kavi
Y Combinator, then a vertical climb.
In the winter of 2024, Tamarind went through Y Combinator's W24 batch. What followed reads less like a graph and more like a ladder pointed straight up: 700% growth in a single year. The model count climbed past 200, spanning antibodies, peptides, small molecules, enzymes and radiopharmaceuticals, with applications from protein-design and binding-affinity prediction to molecular dynamics. In the first month after one launch, 600 users showed up. The seed money came from YC. The Series A, announced February 2026, came from Dimension Capital's Nan Li, who has spent his career watching AI seep into the core workflows of life science and decided Tamarind was the plumbing.
Forbes summed up the ambition in a headline Kavi clearly enjoys: "The Startup Building An Operating System For Biotech AI." An operating system is the right metaphor. You do not admire an operating system. You forget it is there while it does everything. That is the bar Kavi has set: make the most advanced science on the planet so ordinary that nobody remembers it used to be hard.
Why a website beats a breakthrough.
There is a quieter argument underneath Tamarind, and Kavi makes it plainly: a biotech's computational team should not spend its hours re-implementing public models. "The best way to focus energy as a computational team at a biotech company is to work on novel science or AI tooling," he says. Tamarind does the undifferentiated heavy lifting - the inference, the scaling, the glue - so the scientists can do the part only they can do. It is a deeply unglamorous pitch, and that is exactly why it works. The flashy companies promise the molecule. Kavi promises the workbench.
The customer list tells the story better than any slogan. Bayer. Boehringer Ingelheim. Adimab. Mammoth Biosciences. Companies inside Flagship Pioneering's orbit. Roughly a hundred biotechs and counting. In April 2026 Tamarind partnered with A-Alpha Bio to close the loop between AI design and high-throughput wet-lab validation - the dream of computational biology, where a model's guess gets tested against reality at scale. By June, the company had been selected to build and operate the inference infrastructure behind Eli Lilly's TuneLab 2.0, a collaborative AI/ML drug-discovery platform. When a top-five pharma wants someone to run its inference, it does not pick the company with the loudest demo. It picks the one whose pipes do not leak.
Most chemists and biologists are not programmers and can't write code to use these tools.- Deniz Kavi, on the gap Tamarind closes
The builder behind the button.
Kavi's GitHub reads like an honest scrapbook of how a young engineer learns: forks of pix2pix, of TensorFlow's models, of a gloriously titled repo that brute-forces every scikit-learn model with every parameter. Tucked among them is a Turkish-language machine-learning tutorial - "Türkçe Makine Öğrenmesi" - a small flag planted for his roots. (His first name, Deniz, means "sea.") He has been on Twitter since January 2015, long before he had a company to promote, and his feed today is a running commentary on the frontier he sells into: mini-proteins as antibody alternatives, RFdiffusion2's scaffolding tricks, de novo peptide design, the patent questions that AI-designed proteins are about to force on everyone.
The people who know him describe someone genuinely more interested in advancing the science than in counting the money - a founder who treats specialized knowledge as something to be unlocked and handed out, not hoarded. It is a useful disposition for the job he has taken. The promise of AI in biology has never been the bottleneck. The bottleneck has always been access. Kavi looked at that gap and built a bridge plain enough to walk across without a manual.
The timing was not an accident.
Tamarind exists because two curves crossed. The first is the explosion of open molecular models - AlphaFold's 2021 arrival turned protein-structure prediction from a decade-long puzzle into a weekend query, and a wave of successors followed: RFdiffusion for designing proteins that have never existed, MPNN for sequence optimization, GROMACS for molecular dynamics, a growing fleet for binding-pose and thermostability and solubility prediction. The second curve is demand. Pharma woke up to the fact that AI was no longer a side experiment but a core part of the R&D workflow. The models were getting better by the month, and the number of people who could actually drive them was not. Kavi positioned Tamarind precisely in that scissors - between an abundance of capability and a scarcity of access - and let the gap do the selling.
What makes the bet durable is that it does not depend on any single model winning. Tamarind is modality-agnostic by design: antibodies, peptides, small molecules, enzymes, radiopharmaceuticals, all under one roof. When a better structure-predictor ships next quarter, Tamarind adds it to the shelf and every customer inherits the upgrade without changing a thing. The company is not betting on which AI wins the science. It is betting that, whoever wins, scientists will still want to use it without becoming infrastructure engineers first. "That trend will accelerate," Kavi says of AI's spread through drug discovery, "and our goal is to make these tools ubiquitous for all."
Thirteen people, top-twenty pharma.
The most striking number on Kavi's company is not the funding or the growth rate. It is the headcount. A team of roughly a dozen-plus people supports eight of the twenty largest drug companies in the world and a hundred biotechs besides. That leverage is the whole thesis made visible: when you build the right layer of infrastructure, a tiny team can carry an entire industry's undifferentiated workload. Kavi is hiring across software engineering, applications science - protein design, computational chemistry - and go-to-market, but the lean shape is deliberate. An operating system does not need a thousand authors. It needs a few people who refuse to let the abstraction leak.
There is something quietly contrarian in all of it. In an era of biotech startups racing to announce their own miracle molecule, Kavi chose to build the thing the molecule-makers stand on. It is the less photogenic half of the gold rush - the picks and shovels, the rails, the plumbing - and historically it is the half that lasts. He seems comfortable with that trade. The work is invisible by definition: if Tamarind is doing its job, ten thousand scientists are not thinking about Tamarind at all. They are thinking about the disease. That, in the end, is the most Kavi-shaped ambition there is - to disappear so completely into the workflow that the science is all that remains.
Ask him where this goes and the answer is small and enormous at once. "Our goal is to make these tools ubiquitous for all." Ubiquitous is a quiet word for a loud ambition: a world where any scientist, anywhere, can design a drug candidate before lunch and never once think about the machinery underneath. Kavi is betting that the future of medicine will not be invented by the people with the biggest GPU clusters. It will be invented by the people who finally got to stop thinking about GPU clusters at all.