Airdoc

Project background

Airdoc has achieved joint improvement in technical strength through cooperation with university laboratories and other medical institutions in the industry, in addition to independent research and development. However, it still faces some challenges.

On the one hand, the company needs to use a large amount of data for model training, but as data is the core asset of the company, how to achieve data security and controllability is Airdoc’s core demand.

On the other hand, multimodal data analysis, as the core product of the company, also requires a large amount of AI computing resources. If self purchased hardware is used, it will result in significant procurement and operation costs.

Solution features

In response to Airdoc’s core requirements for data security, operational security control, and AI cloud computing, UCloud and the Airdoc technical team have jointly discussed and developed a public cloud based solution, as shown in the following figure:

Firstly, according to different types of accessing users, they are assigned to different VPCs to ensure macro level network isolation. By using bastion machines, micro level permission control can be implemented on cloud servers within different VPCs, granting different permissions to different users. Through techniques such as logging and screen recording, all operations and data flow can be traced back to ensure data security. The combination of bastion host and VPC supports the constantly expanding business needs of users, and while user data is fully shared, its security is also effectively guaranteed.

Secondly, in the selection of medical data storage, considering the needs of external network upload and internal network use, UFS object storage is chosen as the carrier for data sharing and exchange. It not only has the conventional characteristics of high concurrency, elastic scalability, and cross availability access, but also can achieve flexible decoupling of data production and use, improving the granularity of permission security management.

Finally, in response to the AI computing needs of massive data, a combination of UAI series products (including cost-effective GPUs, UAI Train, UAI Inference, etc.) is provided. It is a P40 based cloud host cluster, a large-scale distributed computing platform for AI training tasks, providing one-stop training task hosting services and a convenient deployment environment for users. The relevant drivers and architectures have been configured by default, allowing users to focus only on optimizing model algorithms. The on-demand charging model also provides Airdoc and its partners with a highly stable computing environment, greatly saving the training cost of AI models.

Consumer benefits

Supporting users’ continuously expanding business needs.

While user data is fully shared, its security is also effectively guaranteed.

Flexible decoupling of data production and usage can be achieved, enhancing the granularity of permission security management.

The on-demand charging model also provides Airdoc and its partners with a highly stable computing environment, greatly saving the training cost of AI models.