Tamarind Bio Is Driving AI Assisted Drug Discovery With AWS

Most often life scientists still aren’t able to take advantage of AI models to drive leading research at a fraction of the time. If they are, they must rely on colleagues with tech expertise to deploy and run large language models (LLMs) on their behalf. Not only does this slow down momentum, but they often struggle to keep up with the latest models. CEO and Co-founder of Tamarind Bio (opens in a new tab), Deniz Kavi, explains the problems he witnessed first-hand as a data scientist: “In my undergraduate lab at Stanford, I was quite literally hired to be the human back end running these models. We had so much demand from publications and collaborators that it became untenable to run it manually.”

Realizing that biotechnology and pharmaceuticals companies face the same roadblocks, Deniz Kavi and Sherry Liu co-founded Tamarind Bio in 2023. With a vision to put the power of AI in the hands of any scientist, they developed a platform on AWS to help turn ideas into real-world medicines as quickly as possible. The startup now serves world-leading pharmaceutical companies every day and has helped thousands of scientists discover new molecules. As Sherry Liu, CTO and Co-founder, says: “Scientists have saved thousands of hours they would have spent setting up models, running local scripts, and wrangling data, in exchange for a seamless platform with all the best tools in one place.”

A cure for complex computational biology

Tamarind’s simple interface makes AI accessible for all scientists by taking care of all the infrastructure requirements and connecting them with the latest models as they emerge. R&D teams are equipped to solve society’s biggest medical challenges with large-scale use of advanced models like AlphaFold and RFdiffusion. Tools like these enable teams to better understand what proteins do and how they interact with other molecules in much shorter timeframes—a notoriously complex yet critical component of making ground-breaking discoveries. As Liu explains, “We allow scientists to use these models in pipelines, which otherwise would require custom configuration to set up and keep updated.” Because scientists don’t need to worry about code, Tamarind helps them focus their time and resources where it matters.

As a startup, Tamarind was tasked with winning the trust of established companies before they could showcase the real-world impact. Working with AWS from the outset to build its platform, it was able to kick-start their rapid growth. As Kavi says, “If we weren’t using the most trusted cloud provider, I don’t think it would have been viable to work with large pharma companies.”

The partnership has helped Tamarind quickly build credibility, while also offering hands-on implementation guidance. The business now uses dozens of AWS services to host several hundreds of models and stay agile to changing demands, including Amazon Bedrock (opens in a new tab), Amazon EC2 (opens in a new tab), Amazon S3 (opens in a new tab), Amazon DynamoDB (opens in a new tab), and Amazon EKS (opens in a new tab). “We manage a lot of infrastructure and that is done very seamlessly through AWS. It would not be possible for us to build it on a data center or use another cloud provider in this context,” says Kavi.

Expediting experiments with agentic AI

In its mission to make protein engineering tools simple and accessible, Tamarind is integrating an agentic AI assistant to “run jobs and analyze results,” says Liu. Kavi explains that “Scientists have the intuition and ideas to solve problems, but they often don’t have the expertise to create a system to simulate experiments. We believe agentic AI will enable them to think more about the big picture.”

The startup is stripping away the complexity of computational science by connecting all its AWS workflows and tooling to an LLM on Amazon Bedrock. Previously scientists would have to write code for each task in the drug discovery process, adding everything from ions to waters and membranes one by one. Tamarind Copilot, its agentic AI tool, takes care of this extensive preparation and enables users to simulate complex molecular systems in plain English and iterate experiments in a way that wasn’t possible previously.

Describing the process of developing the tool, Kavi says: “AWS folks guided us through how to best implement and make it scalable. It was a smooth process to implement it from there.” Since the launch, Tamarind can keep developing its agentic AI offering in a structured and secure way as the technology advances using Amazon Bedrock. “Having the new models available in AWS over time is a big help for us. It’s the same standard schemas and the same trusted IT requirements,” Kavi adds.

Scaling scientists’ capabilities

Given the complexity of running test models, a significant amount of compute is required to deliver against spikes in demand. With AWS, Tamarind can take advantage of high-performance scalability without making large upfront investments. “With auto scaling features from AWS, we can go from zero to 10,000 GPUs in a matter of a few minutes,” says Kavi. The startup was able to quickly prove the power of its product to large pharma companies while only using the resources it needs, as it needs them.

To spin up and down hundreds of models on demand securely, Tamarind uses Amazon EC2 and Amazon EKS. “Using spot instances has allowed us to save more than 50 percent on compute, while maintaining scalable availability. Handling our spiky workloads would not be possible without AWS,” says Liu. It also uses Amazon S3 for scalable object storage and Amazon DynamoDB for a fully managed database with high performance at any scale.

Since its early days, dedicated AWS account managers have also supported the business through ongoing guidance, workshops, and AWS Activate Credits (opens in a new tab) to explore different services. “AWS truly cares about our needs and is able to provide technical support and rapid response times,” says Liu. Whether it’s optimizing costs or collaborating on product development, Kavi adds: “One amazing thing is that we can feel that AWS is a very long-term partner. AWS is very focused on making sure we’re successful and we feel very aligned.” From putting Tamarind’s goals first to acting as “extension of the team”, Tamarind’s explosive growth since its inception has been backed by comprehensive expertise and bespoke cloud architecture advice.

Securing world-leading customers

In the biotech and pharma world, data is extremely sensitive, covering unpatented molecules that might eventually become a billion-dollar drug. Any event of a leak or loss of data could therefore have a devastating impact on organizations. As a result, they place the highest importance on upholding strict security standards and intellectual property requirements. As a vendor seeking to process this data, one of Tamarind’s initial challenges was to demonstrate their security credentials to established pharma companies.

Having learned that prospective customers were already using AWS, the business discovered it could quickly build trust through its partnership and streamline procurement. “AWS is a trusted provider for R&D organizations. Most folks ask us to use AWS from a security and scalability perspective,” says Kavi. AWS Key Management Service (opens in a new tab) and AWS Identity and Access Management (opens in a new tab) have also “made it possible for us to create a secure multitenant infrastructure,” says Liu.

As it continues its rapid growth trajectory, Tamarind is seeking to expand its customer base to include smaller research labs and biotech companies. Its strategy involves increasing co-marketing initiatives with AWS, such as building exposure at key industry events and conferences. By doing so, it hopes to become a “one stop shop” for R&D and “the go-to platform where computational drug discovery is done,” says Liu.

Drugging “undruggable” diseases

With an ultimate goal to make R&D more efficient and effective, Tamarind is taking a two-pronged approach. While it continues to pursue new ways of using AI to optimize existing drug discovery programs, it’s also seeking to target diseases that haven’t previously been evaluated or considered. “Because AI behaves so differently than the physical world, we can try to drug the undruggable and unlock this new part of the disease continuum,” Kavi explains.

Preparing for a long roadmap of products, the startup sees AWS as integral to its long-term journey. A core tactic for tackling undruggable diseases will be its ongoing agentic AI developments to increasingly reduce complexity. If a scientist has an idea for what Kavi calls these “interesting targets”, the business aims to help them immediately start working on it and move to testing much quicker—in turn potentially improving millions of people’s lives.

Have you got a big idea that you’re seeking to build and launch? Or maybe you’re a startup looking to scale? Learn how the AWS Activate Program (opens in a new tab) can kick-start success with free credits, technical support, training, resources, and more. Or, if you’re new to AWS, find out how to get started (opens in a new tab) and explore the wealth of support available throughout your journey.