10.25.23-Digital-CoCoPIE

JAY LIU: All right, great. So my name is Jay Liu, and I'm the CEO and co-founder of CoCoPIE. And I'm also a graduate of MIT Sloan MBA class of 2005. Happy to be here. Thank you so much for having us.

We are CoCoPIE.AI. We are not a drink company or cake company. We are an AI software company. And CoCoPIE actually stands for the core technology, which is to optimize next generation large AI models from cloud to the edge.

All right. So over the last decade, AI models have been growing exponentially larger and larger. And they are growing faster than what Moore's law can support. Now with the generative AI, they are getting even much bigger. So that leads to two problems. Number one is the computational cost to support to run those AI models is going to skyrocket. Not sustainable. The other issue is those AI models are too large or too slow to run on edge devices, blocking a lot of very important use cases for edge AI.

So how to solve these problems? Optimization. So we should make the AI models smaller, faster, more cost effective. AI model optimization involves different stages, different phases. AI compression, AI model compilation, runtime optimization. But if you look at today's tool set, this is a highly fragmented toolchain. Different parts do not work or cooptimize with each other. They do not support next generation generative AI models. Using an analogy, it's like trying to assemble your own car in your own garage.

At CoCoPIE, we co-designed the compression-compilation process to maximize performance, intelligence, and energy efficiency. Here you go, CoCoPIE. And once we're able to streamline the whole pipeline end to end to harness the synergy between different parts-- the first and only one in the industry-- magic happens. Now we are able to deliver up to 10 to 50 times speed up across a variety of AI models, from the classical AI models to next generation generative AI models. Now rather than having to assemble your own car, you have a racing car in front of you.

All right. So we have productized our technology in two ways. One, you can use our platform to optimize your own model using your own data, either on prem or in the cloud. And also we have pre-optimized AI models that you can use directly, for example Super Resolution, Stable Diffusion, and some other large AI models.

So let's start with the Super Resolution model. So this is the model that we have optimized so much that you can increase the video resolution in three or four times in real time across a wide variety of end user devices. And the user benefit is that you can significantly reduce the bandwidth cost, the storage cost up to 70%. Also you can delight users significantly by delivering a higher resolution videos in real time.

And this solution has been deployed by one of the largest video streaming platforms in Asia, iQIYI. And also we are official partner with Cognizant, so this solution has been integrated with a joint solution developed by Cognizant, AWS, and Verizon for Smart Stadium.

So another example will be Stable Diffusion. So this is the model that can generate realistic images based on text prompt. So we have optimized the model so much that you can reduce your computation cost in the cloud by more than 10 times. And also we have optimized the model so much that you can deploy the model directly on your mobile devices.

All right. So we are looking for global partners across verticals, whether you are technology, telecom, media, manufacturing, defense, IT consulting. As long as you are using large AI models, probably we can help you. Please stop by. I look forward to conversation with you. Thank you.