Xiaocong Li Email and Phone Number

Recommendation System

- Participated in the development of the advertising data platform (Delivery Forensics). The system processes data generated during advertising, including recall, filter, sort and other data generated by the ad server, as well as client-side beacon data such as impression and completion. Those data can be used to show the delivery status of ads and can also be used to debug abnormal delivery.- Design and implement pipelines： - Consume data from Kafka and Kinesis using Flink to support real-time data querying. - Process offline data using Spark jobs scheduled by Airflow to backfill data after MRC. - Using SingleStore pipeline to inject data and ensure data from real-time job, hourly backfill jobs and daily backfill jobs will not conflict.- Improve Flink Job performance: - Use AWS Lambda Function + Web Service + Local Cache in Flink operator instead of using broadcast to reduce checkpoint size, improve checkpoint speed and stability. - Use flame-graph in Flink job to locate performance bottleneck. - Maintain aggregation status inside an operator and send data to next operator by batch to reduce network traffic and increase throughput.

- Maintain backend business logic of the Toutiao Media Platform (mp.toutiao.com), including Toutiao and Xigua Videos etc. - In charge of developing the querying and publishing services of Unified Content Platform. Content data are stored in MySQL and a KV storage. Those data are also cached in multiple Redis clusters and local cache of Docker pods. The traffic can be more than millions of queries per second.- Lead the development of a general online data storage service. Based on a strongly consistency KV system (ByteKV developed by ByteDance), the storage service can support schema modification, filed indexing, data encryption, auditing and cache, etc.

Some hardware testing and report generation job in the IoT group as an intern