04.10-11.24-HST-Startups-Pyte

SADEGH RIAZI: Awesome-- good afternoon, everyone. My name is Sadegh Riazi. I'm the cofounder and CEO of Pyte. I started the company together with Dr. Ilya Razenshteyn, who received his PhD from MIT in 2017 and was awarded the best PhD thesis.

We are a startup backed by Eric Schmidt's Innovation Endeavors fund as well as National Science Foundation. We started Pyte to make data accessible instantly and globally without compromising on privacy and security of the data.

Essentially, we are a team of cryptographers and technologists that, throughout our careers, we observed frequently that innovative projects are blocked because there is a friction between using data and protecting it. So on the one hand, you want to allow more access for your data scientists and work with more companies. And, on the other hand, you want to keep data confidential and protect it in an isolated fashion. So, essentially, we want to remove this friction.

Let's start by looking at an example, developing a new drug. It takes approximately 8 years and $2 billion. And now the question is that can AI help? Per one study, AI can offer time and cost savings of at least 25%, conditioned on having access to the right data. But privacy regulations pose a huge hurdle. It's even hard to have intra-company data accesses, let alone collaboration with other companies.

To better understand what is the root cause, we need to take a brief look at the architecture of how data collaboration is being done today. A common practice is to simply share data with other research groups or other institutions. Even though data remains secure when you are transmitting it, the other side has to have access to the data, and that creates a lot of privacy and security concerns.

A more recent solution is based on data-sharing environments, like data clean rooms, where you onboard your data into a third party environment, and you run the computation over there. But as you can imagine, not only this requires onboarding data, you need to trust the underlying vendor, and there is no formal security guarantees.

If you mask data, add noise, or remove some part of the data, that's also not effective because, at the end, you have to upload your data to the shared environment, and also, the accuracy of the final AI model would be reduced heavily.

At Pyte, what we have done is that we have created a new data collaboration software, where companies can encrypt their data locally, keep it encrypted, and collaborate with any other research groups or companies without disclosing it. So essentially, each source has their own private key. They encrypt data, and you can collaborate directly over encrypted data without ever opening it.

In fact, we have delivered our software to Fortune 500, and it's already in use. In one case, we helped a healthcare call center to automatically process encrypted patients' requests and classify them without seeing anything about the sensitive data or the PHI that is embedded in that request.

In another case, which actually got its start from MIT as [INAUDIBLE] program, we helped a consumer packaged goods company to securely collaborate with their partners and choose the most compatible scent formula for a variety of their products without disclosing anything about their IP and receiving any information about IP from the other side.

So we are here today to look for any partnerships with companies that have data collaboration or data utilization problem arising from privacy restrictions, security concerns, or IP confidentiality roadblocks. If you face these challenges or if you know anyone within your company who faces these challenges, please come and talk to us. We have a table in the other room. And thank you for your attention.

SPEAKER: Thank you.