“Google Cloud fully meets our infrastructure needs by providing diverse and powerful tools,” he adds. “We used Firebase to quickly launch reliable services at the initial stage, and now GKE enables us to manage and frequently update multiple services at ease. There has been no downtime caused by a lack of computing resources.”

To implement data-driven decision-making, OneClass stores a part of its business data in Cloud Storage and deploys its data warehouse in BigQuery to retrieve data from various sources for analyses, with the results visualized through Looker Studio. As BigQuery supports various data formats, the OneClass team can easily realize speedy cross-database queries by synchronizing data automatically between its databases and BigQuery. Before using Google Cloud, it required several weeks for the team to generate a data analytics report based on different data sources. Now, nearly all analytics requests can be completed within one hour, which has significantly accelerated OneClass’ decision-making process.

Besides, using Cloud Storage has also simplified OneClass’ DevOps work. Many features of its internal management systems and learning platforms require consecutive actions of different services. For example, the recordings of calls made by OneClass salespersons and customer service staff need to be sent from its call center system to Cloud Storage for storage, and then the stored data will be transcribed by Speech-to-Text and analyzed by BigQuery for internal auditing. Cloud Storage not only supports automatic storage of data produced by the servers in Google Cloud but also enables simple trigger pipeline setup together with Eventarc. OneClass’ DevOps team only had to make a few clicks to send triggers automatically between services, instead of spending one to two months on development.

Now, as Gemini can digest and produce texts accurately, OneClass counselors only need to do simple reviewing and editing. This way, they can track students’ learning progress more easily and give more personalized, effective advice. Leveraging Gemini to generate the initial draft of the replies to students’ questions, OneClass tutors have also seen their workloads reduced by 30% and are now able to dedicate more time to lesson preparation.

Besides text generation, OneClass utilizes other AI capabilities of Google Cloud to improve the learning experiences that it offers. It employs Text-to-Speech AI, Speech-to-Text, and Gemini to develop a virtual teacher for English conversation and pronunciation on OneWordz. This means that its users can practice English speaking anytime and receive real-time feedback, which helps them progress faster. By using Imagen on Vertex AI to generate images for English vocabulary flashcards, the OneClass team can produce learning materials five times faster compared to the manual method. 

“Using Vertex AI and other AI tools of Google Cloud, we can now produce quality learning materials and study reports much more efficiently, allowing us to help our students achieve better results more easily and effectively,” notes PinLin Ho, Head of Research and Development at OneClass. “The performance of these tools is industry-leading. For example, by giving simple prompts, more than 90% of the images generated by Imagen 3 are good for us to use. Other similar offerings only reach 30%.”

As its usage of AI models increases, it requires the OneClass team more time and expertise to set up the computing environments that are the most suitable for AI models. To improve its AI model performance with the least DevOps effort, the team in mid-2024 started using Vertex AI Workbench instances to optimize the computing environment of its AI models deployed in Google Cloud. It used to take several hours for OneClass AI models to transcribe and summarize a tutoring lesson. Now, as the Workbench instances provide the most optimal development environment, the same process can be done in five minutes, further enhancing the company’s productivity.