Designed and implemented a sensitive content warning system capable of real-time detection of sensitive behavior in teachers’ audio and video streams, promptly reporting such behavior within a 3-second timeframe.

Implemented the producer-consumer design pattern through message queue to integrate key components of the system, including stream data reception, data preprocessing (frame extraction and audio stream interception), encapsulation of detection tasks (image/audio sensitive content recognition), task distribution, task consumption, and sensitive behavior reporting.

Employed Thrift RPC to facilitate communication with an image recognition service for detecting sensitive content within image frames. Additionally, utilized WebSockets to establish communication with a voice-to-text service, enabling the transmission of audio data and reception of text content.

Created an audio cache using Java’s TreeMap, enabling the system to cache incoming audio, query audio based on timestamp ranges, and automatically clear expired audio data after the detection process.

Designed and implemented an automation system that generates Flink SQL automatically, resulting in an 80% reduction in engineers’ workload.

Developed an independent service using SpringBoot to establish seamless communication with various components, including the metrics configuration platform, big data development platform, and MySQL database.

This service’s primary functions included the automated generation of Flink SQL, creation of big data development tasks, syntax validation of Flink SQL, assignment of reviewers, and publication of Flink SQL Tasks.